EVALUATING THE ROLE OF HYPOGLYCEMIA & COMORBID ILLNESS ON DIABETES MANAGEMENT BEHAVIORS

Over the recent years, the incidence of diabetes has significantly increased. Use of preventive services and treatment with oral antidiabetic drugs (OAD) has remained cornerstone in the management of type 2 diabetes (T2DM). Despite strong evidence that treating diabetes using these disease management strategies decreases morbidity, mortality and complications, glycemic control as well as other diabetes related outcomes remain unsatisfactory. The prevalence of diabetes in the United States has grown vastly in proportion over the last few years with the American Diabets Association estimating that 9.3% of the population suffered from diabetes in the year 2012. These estimates are expected to increase in the future with the World Health Organization (WHO) estimating that 366 billion people (4.4%) will have diabetes. Diabetes places a greater clinical as well as economic burden on the patients as well as the health care system. The presence of comorbid depression is frequent in people suffering from diabetes and can cause health outcomes in patients with diabetes. Poor adherence and persistence to diabetic medications resulting from the occurrence of adverse events is a cause of poor health as well as economic outcomes. There is a continuing need to evaluate the associations between comorbidities as well as common complications of medication treatment in persons with diabetes and examine how they influence health behavior. Evidence regarding differences in the utilization of preventive care services in diabetic patients with and without comorbid depression is scant. Similarly, the factors that predispose an individual to hypoglycemia as well as the association between hypoglycemic episodes and persistence to OAD therapy, specifically sulfonylureas, has rarely been quantified retrospectively. This dissertation utilizes the manuscript format and has four fold objectives: 1. To review the current literature regarding the role of hypoglycemia and comorbid depression in the diabetes and examine their impact on clinical and economic aspects of diabetes management; 2. To quantify the effect of comorbid depression on the rates of preventive care service use in a nationally representative population of US adults; 3. To identify significant predictors and estimate the costs associated with the occurrence of hypoglycemia in the inpatient and outpatient settings. 4. To evaluate the association between the development of hypoglycemia and persistence to oral sulfonylurea therapy in patients newly initiated on this class of OAD medications. In order to review the literature regarding the effect of hypoglycemia and comorbid depression and diabetes, we utilized various biomedical and psychological databases. We analyzed the effect of comorbid depression as a principal risk factor associated with use of ADA recommended preventive services in patients with diabetes using the Medical Expenditure Panels Survey Data. A logistic regression was performed to achieve this objective and all the relevant confounders were controlled for in order to achieve the results. Claims data provided by the Blue Cross and Blue Shield of Rhode Island was utilized to assess the relationship between hypoglycemia and persistence to sulfonylurea medication as well as outline the predictors and costs of hypoglycemia. A time-varying Cox proportional hazards regression model was utilized to compare the hazard rate of medication discontinuation in diabetic patients that were exposed to hypoglycemic events, compared to those that were unexposed. A predictive modelling approach was utilized to highlight the factors associated with hypoglycemia. While the impact of comorbid depression and diabetes was significant both clinically and economically, it was seen that the extent of preventive care service use was comparable for diabetic patients with and without comorbid depression but suboptimal in general thus indicating major gaps in the implementation of ADA recommended preventive care practices. While depression was not significantly associated with increased use of the recommended diabetes preventive care services, other sociodemographic factors were seen to contribute. Moreover, though no significant association between events of hypoglycemia and subsequent discontinuation sulfonylurea medication was illustrated, we demonstrated several clinical factors to have a profound impact on the risk of developing hypoglycemic episodes.

the association between hypoglycemic episodes and persistence to OAD therapy, specifically sulfonylureas, has rarely been quantified retrospectively. This dissertation utilizes the manuscript format and has four fold objectives: 1. To review the current literature regarding the role of hypoglycemia and comorbid depression in the diabetes and examine their impact on clinical and economic aspects of diabetes management; 2. To quantify the effect of comorbid depression on the rates of preventive care service use in a nationally representative population of US adults; 3. To identify significant predictors and estimate the costs associated with the occurrence of hypoglycemia in the inpatient and outpatient settings. 4. To evaluate the association between the development of hypoglycemia and persistence to oral sulfonylurea therapy in patients newly initiated on this class of OAD medications.
In order to review the literature regarding the effect of hypoglycemia and comorbid depression and diabetes, we utilized various biomedical and psychological databases.
We analyzed the effect of comorbid depression as a principal risk factor associated with use of ADA recommended preventive services in patients with diabetes using the Medical Expenditure Panels Survey Data. A logistic regression was performed to achieve this objective and all the relevant confounders were controlled for in order to achieve the results. Claims data provided by the Blue Cross and Blue Shield of Rhode Island was utilized to assess the relationship between hypoglycemia and persistence to sulfonylurea medication as well as outline the predictors and costs of hypoglycemia. A time-varying Cox proportional hazards regression model was utilized to compare the hazard rate of medication discontinuation in diabetic patients that were exposed to hypoglycemic events, compared to those that were unexposed. A predictive modelling approach was utilized to highlight the factors associated with hypoglycemia.
While the impact of comorbid depression and diabetes was significant both clinically and economically, it was seen that the extent of preventive care service use was comparable for diabetic patients with and without comorbid depression but suboptimal in general thus indicating major gaps in the implementation of ADA recommended preventive care practices. While depression was not significantly associated with increased use of the recommended diabetes preventive care services, other sociodemographic factors were seen to contribute. Moreover, though no significant association between events of hypoglycemia and subsequent discontinuation sulfonylurea medication was illustrated, we demonstrated several clinical factors to have a profound impact on the risk of developing hypoglycemic episodes. v

ACKNOWLEDGMENTS
. The completion of this thesis has made possible with the care, support, and encouragement of numerous people including my family, instructors, friends and colleagues. At the end of this journey, I would like to thank all people who made this thesis a possible and an unforgettable experience for me.
First of all, I would like to thank my advisor, Dr. Brian Quilliam for his unconditional guidance, support, patience and encouragement. He has always been like a guardian as well as a constant source of inspiration and I will always remain thankful for having such a knowledgeable mentor and exceptional human being as my major advisor. I will always be indebted to him for providing me with the opportunity to learn from him, which has helped me build a strong platform for my future endeavors. He will always retain a very special place in my life and I look forward to this association growing stronger in the future. I may not have been able to reach where I am today without his invaluable presence. I take this opportunity to sincerely acknowledge all people at the University of Rhode Island who have become an important part of my life in the past few years. I will always remain eternally obliged to Dr. Celia Macdonnell for being a support system over all these years. Working for her has been one of the best experiences of my student life. She has been a guiding light over all these years, preparing me for the future and treating me like family. Dr. Paul Larrat has been one of the most wonderful people I have met and I have always enjoyed speaking to him on various vi topics. I am grateful to Dr. Cynthia Willey for all her inputs, suggestions and attention which has helped me immensely to grow academically ever since I first attended her classes. Dr. Stephen Kogut has been one of the most welcoming and helpful people to me. He has always been approachable whenever I have needed any advice. He has taken keen interest in my development as a PhD candidate and I truly appreciate that. I am highly obliged to have been able to work with Professor Rita Marcoux. She has been a pillar of strength for me over all these years. Apart from wonderful academic experiences as a TA, her support and care has been immense and unwavering. I would also like to say a heartfelt thank you to Barbara Ray, my supervisor for the safety work I have been doing over the years. I will always remember her generosity and concern for me ever since my first days here at URI. I would also like to thank Mr. Chuck Wentworth for his valuable suggestions and motivation during my work with the claims based data. Without their support and input and personal cheering, it would have not been possible for me to complete my degree. I will always remain eternally grateful to this select group of people.
My gratitude is also extended to members of my dissertation committee, Dr.
Rachel DiCioccio and Dr. Anne Seitsinger. They have been extremely kind and helpful to take time out of their busy schedules and attend my defense.

Diabetes
Diabetes is a chronic metabolic disorder of characterized by different pathways. It results in hyperglycemia with an abnormality in the body's capability to convert glucose (sugar) to energy. Type 2 diabetes (T2DM) is the most frequent subtype of diabetes since it accounts for up to 90% of all cases of diabetes worldwide. 1 Diabetes is the seventh leading cause of death in the United States. 2,3 Diabetes is the most significant cause of macrovascular and microvascular complications. 4,5 Patients with diabetes have a lower quality of life as compared to people of the same age group without diabetes and it is even lower in cases of diabetes complications and disease progression. [6][7][8] Recent estimates put the prevalence of diabetes for individuals aged 20 to 74 worldwide at 6.4% in 2010, and it is estimated that the prevalence would increase to 7.7% (439 million patients) by the year 2030. 9 It is one of the most prevalent, debilitating, and costly chronic conditions, both nationally as well as globally, resulting in substantial mortality, morbidity, and economic burden. 4,5,10 During the last 20 years, the prevalence of diabetes has increased dramatically in many parts of the world and the disease is now a worldwide public health problem. The World Health Organization estimates that the total number of people with diabetes is projected to rise from 171 million in 2000 to 366 million in 2030. 11 According to the American Diabetes Association (ADA), the prevalence of diabetes in the United States was about 9.3% in the year 2012 with 29.1 million people diagnosed with this disorder according to the Centers for Disease Control (CDC), which was significantly higher than the 25.8 million (8.3%)reported in previous years. 12,13 In addition to this, another 7 million people are estimated to be suffering from undiagnosed diabetes and 86 million are estimated to suffer from prediabetes. 14 The CDC estimates that among U.S. residents aged 65 years and older, 10.9 million, or 26.9%, had diabetes in 2010. 15 This number is projected to rise rapidly with the ageing of the population and the corresponding increase in the prevalence of chronic conditions. 2,16,17 Danaei et al. in a study to estimate the undiagnosed diabetes prevalence as a function of a set of health system and sociodemographic variables, found that prevalence of diabetes in the U.S. was 13.7% among men and 11.7% among women ≥ 30 years. 18 Previous studies have noted major cultural and racial variations in the prevalence of T2DM. 13,19,20 According to the ADA, 12.6% non-Hispanic blacks have T2DM compared to 7.1% non-Hispanic whites in the United States, with the highest prevalence of diabetes found in the southern states. 13,21 In a recent study by the ADA to estimate the total economic cost of diabetes for the United States, it was demonstrated that in the year 2007 the economic burden was estimated to be $174, approximately 1 of 8 dollars spent on medical care, as compared with $98 billion in 1997. 10,22,23 In the same year, the approximate total cost for treating diabetes was $232 billion. Moreover, the total costs of diabetes had increased to $245 billion with $176 billion attributable to direct medical costs and $69 billion to lost productivity. 10 After adjusting for population age and sex differences, average medical expenditures among people with diagnosed diabetes were 2.3 times higher than those in the absence of diabetes. 10 Literature suggests that a high degree of health care resource use can be attributed to diabetes namely hospital inpatient days (25.7%), nursing/residential facility days (33.3%), prescription medications, and visits to the physician, emergency room, hospital outpatient etc.
Previous research suggests that these estimates are higher in cases of uncontrolled diabetes and diabetes with complications. 10 The increasing prevalence of the disease and thereby its economic as well as social impact emphasizes the importance of effective diabetes prevention and care.

Use of Preventive Services in Diabetes
Apart from being a major cause of heart disease and stroke, diabetes poses an increased risk of cardiovascular, peripheral arterial and cerebrovascular diseases. [24][25][26] Previous studies also note a much higher proportion of non-traumatic lower limb amputations, kidney failure and blindness in diabetic patients. 26 A number of studies have concluded that timely utilization of medical and preventive care is an ideal practice for the management of diabetes. 27,28 These services can be vital in the incidence and progression of any diabetes specific complications. 29 Controlling blood glucose, blood pressure, and LDL cholesterol levels can reduce the microvascular complications like eye, kidney, and nerve diseases as well as macrovascular complications like heart attack, stroke, and lower-extremity amputations. For example, Litzeman et al. in a blinded, randomized, controlled trial of patients in an academic general medicine practice setting found that better foot care in patients with diabetes were likely to have reduced prevalence of lower extremity clinical disease. 28 One of the principal preventive services in patients with diabetes is the testing the A1C which might demonstrate the patient's blood glucose levels.
Similarly, routine eye exams diagnose symptoms of diabetes related eye disease and this early detection is usually instrumental in preventing the progression this disease.
Comprehensive foot care programs include assessment and treatment of feet of the diabetes patient and can help reduce amputation rates. Other preventive tests include timely immunization against pneumococcal disease and influenza as well as regular blood pressure and medical checks.

Depression
Major depressive disorder (MDD) is a grave and recurrent condition affecting around 121 million people worldwide. The World Health Organization (WHO) recognizes depression as the fourth leading cause of disease burden associated with non-fatal health outcomes and a leading cause of disability around the world. 30 It has been widely reported in the literature that depression is often underdiagnosed and under treated. 30,31 Even though the scenario with respect to treatment rates has been positively changing in the past decade, many patients still suffer from symptoms of depression. 32 Gonzalez et al. in a study of a large national sample found that majority of the people suffering from depression did not get the guideline recommended degree of care and large disparities existed based on various factors like race and ethnicity. 33 Kessler et al. in a study using the National Comorbidity Survey Replication (NCSR) suggested that in the United States, the lifetime prevalence of depression was 16.2% and the 12-month prevalence was 6.6%. 32 Bromet and colleagues, in a study conducted in in over 18 countries reported that the average lifetime prevalence of depression was 14.6% in high-income countries and 11.1% in middle to low income countries. 34 Mathers et al. in a study projecting mortality and burden of disease by cause to year 2030 found that depression was predicted to hold the second position among diseases contributing to the global burden of diseases by 2030 , and there has been a 37% increase in disability-adjusted life years of depression from the year 1990 to 2010. 35 Evidence from studies suggests apart from being a major cause of morbidity, mortality and disability, MDD is responsible for higher economic costs by means of health care resource use as well as indirect costs including workplace absenteeism, diminished, or lost work productivity and increased use of healthcare resources. 6,[36][37][38] This economic burden of depression has been evaluated in several studies nationally as well as worldwide. For example, Greenberg

Comorbidity of Diabetes and Depression
A large body of literature has highlighted the association between diabetes and depression. [46][47][48] The essential finding in the aforementioned analyses has been that that diabetes and depression co-occur frequently with the presence of once condition significantly increasing the likelihood of patient suffering from the other. 48,49 A systematic review by Roy and colleagues suggested that prognosis of comorbid depression in patients with diabetes, in terms of its clinical and societal implications, is worse for either condition in comparison to when they occur individually. 50

A Bi-Directional Relationship between Diabetes and Depression
In a review of literature Egede et al. noted that previous research indicated a complex relationship between diabetes and depression because the temporality of this association is not clear. 6 There is a growing body of evidence, which suggests that there exists a bidirectional relationship between these two chronic disorders. 51 Moreover, several physiological and behavioral mechanisms have been studied to explain this possible link between diabetes and depression. Mezuk et al. in a metaanalysis of studies from 1950 to 2007 of diabetes and depression, established that people with depression had a 60% increased risk of developing diabetes compared to non-depressed patients while people with diabetes had a 15% increased risk of developing depression compared to non-diabetics. 52 Pan et al. studied women over a 10 year period and found that the relative risk for T2DM in women with depressed mood was higher as compared to those who were non depressed (OR 1.17; 95% CI 1.05-1.30). 53 Studies have found incidence of depression as a modifiable independent risk factor in the onset of diabetes. 54-56 Golden et al. performed a longitudinal study depressive symptoms at baseline were associated with an increased incidence of T2DM at follow-up over a 3-year period; an increased risk for developing depressive symptoms over the 3-year period was associated with treated T2DM, but conversely baseline impaired fasting glucose and untreated T2DM were associated with reduced risk for depression. 51 The increased risk of developing diabetes might be due to the negative physiologic effects of depression on glucose metabolism as well as neuroendocrine and autonomic nervous systems. 54 Similarly, diabetes can also act as a precursor to depression through various clinical and psychological mechanisms. 50 Poor metabolic control and increasing complications as a result of diabetes may result in or further worsen depression and lessen response to antidepressant treatment. 6,50 For example In a study of 1,586 older adults from the Rancho Bernardo study, Palinkas et al. reported that there was a 3.7-fold increase in odds of depression in those with a prior diagnosis of diabetes. 57 Thus, as suggested in a recent meta-analysis, there is lack of concrete information supporting the direction of relationship between diabetes and depression with there being clinical and epidemiological support for both hypotheses. 58

Prevalence of Depression in Diabetes
It is well documented that depression is significantly higher in patients with diabetes as compared to the general population. 48 8). 67 Ali et al. also found that the prevalence of depression was significantly higher among patients with T2DM (17.6%) than those without diabetes (9.8%). 48

Clinical Implications Of Comorbid Diabetes And Depression
The clinical implications of occurrence of comorbid depression in patients with diabetes are of serious concern since it is associated with poorer diabetes outcomes. 48,68 Apart from having a negative impact on the physical, mental and social wellbeing of the diabetic patients, it might also have implication on quality of life, rates of mortality and morbidity. 49 ($3,264) than patients with diabetes alone ($1,297). They also found that depressed patients with diabetes had higher total medical costs ($19,298) than patients without depression ($4,819). 6 Egede et al. found that adults with diabetes and depression were more likely to miss more than 7 workdays in any given year. 85

Self Care Behaviors In Patients With Diabetes And Comorbid Depression
Self-care behaviors are of highly critical in the management of diabetes since patients with comorbid depression and diabetes are at a higher risk for worse health outcomes as compared to individuals with a single disorder. 86

Oral Hypoglycemic Agents And Goals Of Therapy
Previous research has identified some of the principal goals of oral antidiabetic therapy. ADA proposes a glycated hemoglobin (A1C) level of less than 7% and preprandial blood glucose level of 80-120 mg/dL, a bedtime blood glucose level of 100-140 mg/dL. 90,91 Garber et al. and Grant et al. have

Hypoglycemia in T2DM
Hypoglycemia is one of the most common as well as dangerous side effects of T2DM therapy. Even though it is well proven in evidence that tighter glycemic control may be instrumental in reducing the risk of other serious complications of T2DM like retinopathy, neuropathy, and nephropathy, it can also present an additional risk of severe hypoglycemia. In many T2DM patients, hypoglycemia is responsible for recurrent morbidity and at times, can be a cause of mortality thereby creating a barrier to the long term benefits of optimal glycemic control. 103  Study (UKPDS) reported that 2.4% of patients using metformin, 3.3% of patients using a sulfonylurea, and 11.2% of patients using insulin reported incidents of severe hypoglycemia requiring medical attention over the 6 years of follow up period with hypoglycemia becoming a limiting factor to glycemic control over a period of time. 106

Persistence to OADs in T2DM
Medication persistence or conforming to the recommendation of continuing treatment for the prescribed length of time is an important issue in the long term management diabetes considering the chronic nature of the disease and the nature of the treatment regimen intended to achieve the desired glycemic targets. 107 Cramer et al. in a review of literature to determine the extent to which patients fail to comply with the doses of medications prescribed for diabetes noted that the rates for treatment persistence ranged from 16 to 80% when the patients continued taking their medications for at least 6-24 months. The authors noted that the methodologies followed by the researchers varied in that the cross-overs to an alternative OHA or insulin might not have been counted as discontinuation. 108 Bocuzzi et al. conducted a retrospective analysis of a large administrative pharmacy claims database, using data on continuously pharmacy benefit-eligible members prescribed OHAs, reported that the 12-month persistence rate for the OHA cohort was low, ranging from 31% for alpha-glucosidase inhibitors to 60% for metformin. 109 of the patients discontinued their therapies in 12 months. 108 Other past studies have reported persistence estimates for oral antidiabetic medications from a low of 15% to a high of 76% with the variations attributed to differences in methodologies, definitions of persistence as well as length of the follow up periods. Non-persistence with therapy most often leads to failure in meeting glycemic goal thus leading to avoidable undesirable adverse health outcomes. 111 22

Conclusions
It is clear from the above literature review that issues with diabetes management are highly prevalent and are significantly associated with negative health and economic outcomes. The coexistence of diabetes and depression also results in compromised self-care behaviors which are essential in the management of both the diseases especially diabetes. Of greater importance among the self-care behaviors is the suboptimal utilization of preventive care services since these are particularly essential in preventing the complications in diabetes. Similarly, medication persistence and hypoglycemia are issues affecting the achievement of desired glycemic goals. There is a need to develop strategies to address both patient and other health care related factors in order to increase the potential effectiveness of disease management in diabetes. A multifactorial approach might be essential to counter the adverse health effects of comorbid diseases as well as adverse effects of diabetic medications in patients with diabetes.

Background and Significance
The American Diabetes Association (ADA) publishes Clinical practice guidelines for persons with diabetes annually. These guidelines are evidenced based and are intended to improve the quality of care for patients with diabetes. 112 Evidence from previous research has been instrumental in the development of several clinical practice guidelines that are intended to improve the quality of care for diabetes patients 112 . These recommendations suggests that adherence to these recommended guidelines is likely to have a positive impact on the morbidity and mortality related to diabetes thereby reducing the clinical as well as economic burden of diabetes as well as improving the management of diabetes. 12,27,28,113,114 Despite this knowledge, the use of the clinical preventive services in the U.S. adult population is suboptimal and is quite variable. 115 For example, The Healthy People 2020 initiative, reported low levels of use of multiple clinical preventive services for diabetes as well as other diseases. 116 Previous research suggests the use of preventative services ranges from 10% to 85%, depending on the particular service. 117 Evidence from population-based studies in various settings also indicates that there is a significant difference between reported levels of use of preventive care practices among people with diabetes and the degree of recommended use for optimal level of care. 118 This discrepancy might be primarily attributed to the effects of race/ethnicity, income, health insurance coverage, and comorbidities. 119

Study Design and Data source
To achieve the aims of our project, we conducted a retrospective cohort study among adult patients ( To achieve the aims of the MEPS, AHRQ employs an overlapping panel designs (Appendix 2.2) approach to identify preliminary contacts followed by interviews during five separate in-person rounds with purposeful oversampling of certain groups (e.g., low income, racial minorities). 138  Since we were primarily interested in assessing the use of preventive care in patients with comorbid depression and diabetes the preventive services for diabetes, we specifically utilized the Diabetes Care Survey (DCS) supplement. 143 The DCS a special self-administered paper-and-pencil questionnaire fielded Rounds 3 and 5.
Households received a DCS based on their response to question regarding a health professional communicating with the individual that he or she has diabetes. The DCS specifically collects information regarding the diabetes related medical care that is received by the patients, including medications and previous diagnostic tests including the receipt of preventive services by the individual in the recent past.
Linking this survey, administered once a year over two years, with the household components enables a complete examination of the diabetes related health care received by the patient. 136

Study Cohort
To gain adequate sample size and construct an analytical study cohort for our analyses, we selected data for individuals from households in Panels 12,13,14,15,16 coinciding to MEPS survey years of 2008-2011. Data for each year was constructed based on six rounds of interviews, rounds 1-3 for the panel that was initiated that year and rounds 3-5 for the panel that was initiated in the previous year. Information attained in each round of the interviews pertains to a specific frame of time known as reference period. Thus, the study reference period begins from 1 st January 2008 for panel 12 round 5 and ends on 31st December 2011, which is the end of the last reference period. Table 2.1 provides an overview of the panels and rounds from which the study cohort was selected with respect to the year. Within these included years, we focused on all respondents aged 18 years or older, who self-reported a diagnosis of diabetes with a positive sampling weight.

Cohort Eligibility-Identification Of Diabetes
We defined the inclusion criterion for the study sample as the presence of diabetes in individuals, with or without comorbid depression. We identified diabetes-using patient self-report in answering to survey question regarding diabetes, where respondents were asked, "Have you ever been told by a doctor that you have diabetes ?". 143 If the individual responded affirmatively, we considered them eligible for entry into the final cohort. Individuals who were aged 17 years or younger and those who responded "not sure," "don't know," "refused" or "missing." were considered ineligible 59 . Similarly, we excluded cases of gestational diabetes. During the process of this survey, every person who reported to have received a diagnosis of diabetes was further asked to complete the DCS. In previous studies, Richard et al. and Dismuke et al. have utilized similar procedures to recognize diabetic population in this dataset to investigate the racial disparities in the quality of diabetes care and association between major depression, number of depressive symptoms and personal income among the diabetic population respectively. 68,139

Exposure Definition-Comorbid Depression
To conduct our retrospective cohort study, we created two groups, 1) those with comorbid depression, and 2) those without comorbid depression. We identified study participants with depression using individual's self-reports as well as diagnosis codes captured on the MC survey. As part of the MC survey, the two-item, Patient Health Questionnaire (PHQ-2) is collected from participants. The PHQ-2 is a previously validated instrument to identify depression. The PHQ-2 was designed to report the feelings of depressed mood and anhedonia in patients over the past 2 weeks, with the scores ranging from 0 ("not at all") to 3 ("nearly every day") for each The prevalence of depression in the final study cohort was 25.8% with 1,208 patients suffering from depression as identified by the self-reported PHQ-2, ICD 9 CM/CC codes or both. Of the 4,668 identified respondents, 731 had a score of 3 or higher on the depression scale. Of these, 288 had an ICD 9 CM/ CC code for depression while 443 respondents did not record an ICD 9 CM/ CC code for depression. Out of the 3,937 respondents who did not have a score of 3 or higher on the depression scale, 477 respondents (12.1%) had an ICD 9 CM/ CC code for depression and therefore we classified them as having comorbid depression. Thus, the final sample for inclusion included 1,208 persons with depression and 3,460 persons without depression.

Outcome: Use of Preventive Services in the Past Year
The main outcome variable of this study was the suggested use of diabetes preventive care services according to the ADA guidelines. The self-reported receipt of seven recommended diabetes-specific preventative services was examined within the past year using the DCS, based on the inclusion in the MEPS and consistent with the guidelines of the national organizations. 91 For this study, the seven outcomes of focus were annual receipt of: individuals were asked about the number of times a doctor, nurse or other health professional checked their blood for A1C in the past year. In contrast, questions with reference to the remaining preventive care services were constructed in a manner which asked the respondent to report the last time they underwent a preventive service in the recent past (alternatives restricted to same year, past year, year before past year, not in the past 2 years and never had the preventive service). Using this information, we created seven binary outcome variables for each of the self-reported preventive health care utilization over the prior year. Table 2.2 summarizes the MEPS questions asking about use of these services, the coding scheme for the receipt or no receipt of these services and the ADA recommendations for the services. 91 During the study, respondents with missing data reported for any of the preventive services were considered ineligible for further analyses.

Identification of Potentially Confounding Factors
Based on the available literature, we identified and evaluated a set of sociodemographic characteristics that was associated with differences in the use of preventive care services in patients with comorbid depression and diabetes. The variables describing demographic status of the person were constructed as per the self-reported status of the person on the 31st of December of the survey year. These variables are updated in every round of data collection. Since the DCS was a crosssectional survey, the variables represented current status only. In our study analyses, some categories of the variables were merged and reconstructed to have a preferable distribution of the population within these categories.

Identification of Comorbidities
Based on the available literature, we identified and evaluated a set of sociodemographic characteristics that was associated with differences in the use of preventive care services in patients with comorbid depression and diabetes. The variables describing demographic status of the person were constructed as per the self-reported status of the person on the 31st of December of the survey year. These variables are updated in every round of data collection. Since the DCS was a crosssectional survey, the variables represented current status only. 15  3 provides more detailed information on the identified comorbidities including the CCCs that utilized to ascertain each of the comorbidities.

Statistical Analysis
As we conducted a retrospective cohort study, we conducted all analyses comparing the two created exposure groups, those with comorbid depression (n= 1,208) and those without comorbid depression (n= 3,460). We first checked for comparability in background characteristics of the two groups focusing on age, gender, comorbidity conditions, and other aforementioned socio-demographic factors. This was done to understand and profile the study population and identify potentially confounding factors between the two groups. We generated descriptive statistics within each of the groups including means and standard deviations for continuous variables and percentages were reported for categorical variables. We examined differences between the two exposure groups using the Pearson's chi-square test or Fisher exact test for categorical variables and T tests for continuous variables.
After baseline comparison of the two exposure groups, we next assessed the prevalence of use of preventive care service use in the diabetic population with and without depression. Percentages were reported to compare the use of preventive services between diabetic individuals with and without depression. To further quantify the effect of comorbid depression on use of preventive services after controlling for confounding, we employed logistic regression modeling. We identified potentially confounding variables that varied between the two exposure groups (during bivariate analyses). In addition to these, variables of known clinical importance (age, gender race/ethnicity) were selected as potential confounders.
Initially, we constructed unadjusted odds ratios for use of preventive care services in diabetic patients with and without comorbid depression. Then, we used a series of multivariate logistic regression models to determine the independent effect of comorbid depression on the use of preventive services. Controlling for all the aforementioned patient level covariates that might influence the use of preventive services, we calculated the adjusted odds of receiving at least 2 A1C tests, a diabetic foot examination, an eye examination, an influenza vaccination, a blood cholesterol check and a routine medical checkup in the past year. Because of the complex survey design of the MEPS HC file, we used special diabetes weights from MEPS to compute robust standard errors of the estimates. 150 During the process of building the logistic regression models for each mentioned preventive care services individually, we used a non-computer generated stepwise approach. Variables that were identified as potential confounders in the bivariate analysis were added to the model for each service sequentially in the order on the largest significant difference between the depressed and non-depressed groups with respect to the particular variable. Thus, nested models were fitted in an iterative, manual process using an inclusion threshold of a 10% change in the β estimate of the principal independent variable, comorbid depression (indicator of exposure group).

Final Cohort
Within the four years of MEPS data utilized for our analyses (2008-2011), we identified 138,030 survey respondents included in the dataset. Figure 4 describes the steps that were followed for the selection of final cohort of respondents. As per the recommended data estimations procedures, we did not consider respondents with non-positive person level weights for further analyses since only data for persons with a positive person-level weight can be used to make estimates for the civilian no institutionalized U.S. population. Further, we restricted the population to respondents who were above 17 years of age, who had responded "yes" to the question regarding diabetes and had a positive diabetes weight, which adjusts for DCS nonresponse and weights to the number of diabetics in the US civilian noninstitutionalized population. This selection process resulted in the initial sample of 7,780 respondents with diabetes. Out of this population, only the respondents who had information on all the required variables were selected for the study. Thus, the final sample for the study comprised of 4,468 respondents. This sample size represented 60% of the population that was eligible for the study.
The prevalence of depression in the final study cohort was 25.78% with 1,208 patients suffering from depression according to the self-reports of PHQ-2 or ICD 9 CM/CC codes or both (Figure 2.1). Moreover, 288 patients had depression according to the PHQ score as well as ICD 9 CM/ CC code for depression while 443 respondents did not record an ICD 9 CM/ CC code for depression even though they had a PHQ score 3 or more. Out of the 3,937 respondents who did not have a score of 3 or higher on the depression scale, 477 respondents had an ICD 9 CM/ CC code for depression and were considered as suffering from comorbid depression.

Demographic and Clinical Characteristics of the Cohorts of Individuals With and
Without Depression  vs 54.0%) and annual routine medical checkup (88.6% vs 90.6%), people without comorbid depression had a higher rate of receipt of these services.

Multivariable Modeling-the Effect Of Comorbid Depression On The Receipt Of Diabetes Preventive Care Services.
Crude and adjusted odds ratios with the 95% confidence intervals from multivariate logistic regression models describing the associations between presence of comorbid depression in diabetes and receipt of the preventive care services are described in preventive services that were explored, these differences between people with and without depression were not significant after controlling for various sociodemographic and clinical factors.

Discussion
Using a nationally representative probability survey, we performed a study that had 2 principal goals. The first was to examine patterns of preventive care service use among individuals with and without diabetes using data on non-institutionalized civilian US population. The second was to examine the relationship between the status of comorbid depression and receipt of diabetes specific preventive health practices. This study advances the current body of literature regarding the differences in the quality of preventive care in diabetic patients with and without comorbid depression by exploring the impact of depression as an independent factor on receipt of these services. have shown some of these factors to be associated with odds of receipt of a diabetic foot exam annually. 12,15,122,138 On the other hand, our results were in contrast to some studies that did not find a significant association between these sociodemographic characteristics and use of preventive services. 118,139,161 In both unadjusted and adjusted analyses, depression was not significantly associated with receiving an annual lipid exam. In adjusted analyses, we uncovered associations between this preventive care service and some independent variables like age, race and ethnicity, insulin use as well as co-existence of hyperlipidemia. Various studies have found age to be significantly associated with the receipt of an annual cholesterol screening. 12,15,138,162,163 With respect to the relationship between race/ethnicity and the likelihood of undergoing an annual lipid exam, our results are corroborated by some studies 12,163 but are in contrast to results of others. 15,138 Although the effect of depression was insignificant on the likelihood of receiving an annual influenza vaccination, we found that in adjusted analyses age, race/ethnicity, insulin use, and comorbidities were significant predictors of a flu shot. Our findings suggesting that other factors (apart from depression) may be related to the lack of receipt of preventive services rather than depression itself. These findings are consistent with previous reports that provide an evidence of a strong association between these factors and receipt of the preventive service. 119,138,163 After we controlled for the effects of all other sociodemographic covariates, depression was not found to significantly affect the receipt of an annual retinal eye examination. The findings of this study are unique and add to the body of literature regarding the impact of depression on diabetes specific preventive care services. It confirms that in a large population of non-institutionalized patients with diabetes, the overall rates of receiving diabetes specific preventive care services are sub-optimal. Though the selfmanagement of diabetes has been widely considered as having a beneficial effect on control of the disease, we found that major gaps in the use of ADA recommended preventive care practices persisted. 118 Our findings suggest that, as reported earlier, there is a need for policy makers and physicians alike to place greater emphasis on diabetes preventive care practices. 59 Many patient related factors contributing to 63 differing levels of health care use and overall health of patients were found significant in our analyses. Our findings could be vital in the management of diabetic patients with depression since it will allow researchers to focus on specific action areas that need greater importance and attention since they could affect these selfcare activities critical to health outcomes. 59 It also highlights an insignificant association between presence of comorbid depression and adherence to ADA recommended levels of diabetes preventive care. Patients that suffered from depression in our study, though statistically insignificant, were marginally more likely to receive these recommended tests. This might be attributed to the higher frequency of visits to the physicians, since most of these are physician-initiated activities. 164,165 Age, racial differences, insulin use, socioeconomic factors and access to care measured by presence of primary care physician emerged as principal factors related to use of diabetes specific preventive care services.

Limitations And Conclusions
The interesting findings of this study should be viewed in context of some study limitations. Primarily, one limitation was the use of self-report in diagnosing both diabetes and depression and for identifying use of studied preventive services. As with all the observational studies that utilize self-reports in their design, the study had potential for recall bias. in better health outcomes as measured by these quality indicators. 123 Undiagnosed diabetes as well as severe diabetes is known to be critical factors that affect diabetes care and outcomes. It was not possible to account for these factors. 161 The issue of surveillance bias and measuring the disease severity were beyond the scope of the study. In case of this bias, stratification is often seen as a remedial measure.
In conclusion, our study provides valuable insights into the differences in use of diabetes preventive services with evaluation of the effect of depression as a comorbid condition. The extent of use of preventative services was comparable for diabetic patients with and without comorbid depression but suboptimal overall.
Many of the factors that were found significantly associated with the use of preventive services are modifiable and hence strategies and interventions focusing on these could improve the outcomes in diabetes patients. This data also demonstrates the need to study effective management of depression in diabetic patients since depression potentially affects various self-care activities. Future research should focus on the underlying causes of this suboptimal use of preventive services as well as establishing a causal relationship between depression and selfcare behaviors.

Background
Hypoglycemia is a serious complication that is associated with the treatment of diabetes resulting in a significant burden to the patients with diabetes. 166 While most hypoglycemic events are mild and self-managed, more severe hypoglycemic events require medical assistance and result in the development of serious complications. It is well documented that despite the variations in severity, hypoglycemia is known to cause negative health outcomes including increased morbidity, decreased quality of life, and rare occurrences of mortality in patients with diabetes. 167 The most common symptoms associated with hypoglycemia include palpitations, trembling, sweating, hunger, and confusion. 168 Long-term consequences of hypoglycemia include weight gain, cardiovascular diseases, and coma. 167 Even though the symptoms and complications of diabetes differ among patients, a great degree of decline in cognitive and motor function as well as hormonal counter regulation has been previously documented. 169 The fear of severe hypoglycemia requiring clinical assistance can seriously compromise the self-management of diabetes thereby causing the patients to prefer sub-optimal blood glucose control over incidents of hypoglycemia. 170 Though the estimates regarding the incidence of hypoglycemia in Type 2 Diabetes (T2DM) are varied, previous studies identify several factors associated with hypoglycemia. 167,168,[171][172][173][174] For example antidiabetic medications, particularly insulin and sulfonylureas (SUs), are among the principal risk factors for developing hypoglycemic events. [175][176][177][178] This is concerning as intensive therapy with antidiabetic drug agents is strongly associated with improvement in diabetes control including reducing the risk of developing micro-and macrovascular complications. 95,179,180 Similarly, many patient specific factors are also associated with the development of hypoglycemia. These factors include age and gender as well as physiological factors (e.g. chronic kidney disease and liver disease). 177,181 Lastly, certain behaviors like continuity of physical exercise, intake of food and consumption of alcohol also increase the risk of hypoglycemia. 168,173 Apart from the clinical impact, hypoglycemia has also been shown to pose a significant financial burden to the patient as well as the health care system 167,177,182 . from these studies often lack generalizability. As the incidence rates of people suffering from diabetes will almost double by 2050, 2 more research on this devastating complication of antidiabetic treatment is warranted. Hypoglycemic events will continue to place a greater strain on the health care costs and resources.
To expand on previously published studies, we conducted a cross sectional study of persons with T2DM using an insurance claims database to identify predictors and outline the costs of hypoglycemia treated in an outpatient or inpatient setting. 86

Datasource
For the purpose of our study, we used the Blue Cross and Blue Shield Of Rhode Island

Research design and study population
Using the medical claims, inpatient and outpatient, as well as prescription claims, we conducted a cross sectional study. All the patients included in the dataset had a diagnosis of diabetes. As our interest was in factors associated with hypoglycemia in T2DM, we further excluded patients who had at least one claim for type 1 diabetes (ICD-9 250.X1 or 250.X3) or gestational diabetes (ICD-9-CM 648).

Identification of Hypoglycemia
The principal objective of our study was to identify independent predictors of hypoglycemia in T2DM. In order to identify hypoglycemia in the inpatient and outpatient medical settings, we used an algorithm designed by Ginde and  187 As all episodes of hypoglycemia identified in our study required medical intervention, we considered these events as serious and thus will utilize this terminology in the presentation of our results. Using this information, we created two groups, those with at least one serious hypoglycemic event (n=1243) and those who did not have a serious hypoglycemic event (n=28,128).

Identification of Independent Predictors
We assessed the relevant demographic characteristics of the sample population including age (18-34, 35-49, 50-64, 65 and above) and gender. We used the Charlson's Comorbidity Index in order to examine the composite impact of the burden of comorbid diseases on the risk of having a hypoglycemic event. In addition, we also defined the prevalence of specific individual comorbidities predictors of hypoglycemia using the Elixhauser Comorbidity Index. Previous studies have identified other diabetes micro-and macrovascular complications that might be predictive of hypoglycemia in patients. We also identified the presence of these complications as well as other diseases like influenza and pneumonia, which can potentially to increase the likelihood of a hypoglycemic event. 183,188,189 In case of an overlap of the conditions, we considered the condition defined under one set of comorbidities preferably in the order of elixhauser comorbidity index, micro and macrovascular complications and other diseases.

Characterizing The Use Of Medications
We use a combination of National Drug Codes (NDC) and drug product names in order to identify the use of both diabetic and non-diabetic medications that might have an association with the incidence of hypoglycemia. Among the antidiabetic medications, we specifically identified the use of all the major classes of oral  173,[190][191][192]

Estimating Costs
In order to examine the medical costs for hypoglycemia, we used the total allowed amount paid for the services for both inpatient and outpatient encounters. In addition to the measurement of total costs, we also stratified the costs into 3 mutually exclusive groups namely costs related to hypoglycemia as identified using the Ginde algorithm; 187 costs related to other diabetes-related claims as identified by primary ICD-9 250.XX); and costs related to all other claims. We classified all the episodes occurring on the same day as a single episode of care. All the costs were adjusted to 2012 equivalents (final year of available data) using the regional Consumer Price Index medical care expenditure category in order to make accurate comparison of costs across all study years.

Statistical Analyses
We created two groups for comparison in our study, those with a serious hypoglycemic event and those without. We compared the prevalence of the selected covariates among the patients with or without any hypoglycemic events by examining the frequencies and thereafter using the Cochran-Mantel-Haenszel statistic for categorical variables and Student's t-test for continuous variables. We then selected all the variables with a p-value of less than 0.25 in these preliminary bivariate analyses. We then developed a predictive logistic regression model using the variables identified in the above process, initially fitting a preliminary model containing all of the above variables and then further refining it using a manual iterative process of refinement. During this process, we sequentially excluded variables that were not contributing significantly to the model (Wald p-value >0.10) and thus were potentially not associated with hypoglycemia. Further, we carried out likelihood ratio testing in order to confirm the exclusion. After identifying a working model with all of the relevant predictors included, we further assessed multicollinearity in the model. For this purpose, we used the variation inflation factor (VIF), and Eigen values to make the decisions on exclusion of collinear variables.
These were calculated utilizing a separate regression model and specifically using the VIF, TOL, and Collin options. If two variables were found collinear, we included the variable that was clinically more relevant to our analysis. We tested all two-way interactions between the independent variables in a stepwise process that was similar to the one used to in order to build the initial model (using likelihood ratio testing for confirmation). We retained each interaction term if it was significant and continued this process until all interaction terms were either removed from the model, or retained if found significant. We used AIC (the Akaike Information the final model. At the end of this process we then reported multivariable (adjusted) odds ratios (AORs), including their respective 95% confidence intervals (95% CI). For the cost analyses, we conducted Student t-tests to compare mean costs across the created cost subgroups. All statistical tests were conducted with a 2-tailed alpha of 0.05. All analyses were performed using SAS software version 9.3 (SAS Institute Inc., Cary, NC). This study was reviewed and approved as exempt by the University of Rhode Island's Institutional Review Board.

Demographics
The initial dataset of patients with diabetes was compromised of 36,954 individuals identified as having diabetes  Moreover, the observed gender distribution was similar across both groups with the proportion of females being lower in both hypoglycemic (567 patients, 45.6%) and non-hypoglycemic groups (13,275 patients, 47.2%)

Clinical Characteristics
Comparison of clinical characteristics in patients with or without serious hypoglycemia revealed a higher prevalence of comorbidities in the hypoglycemic group (Table 3.2). Overall, the Charlson's comorbidity score describing the burden of other diseases was found to be higher in the hypoglycemic group (mean = 2.97 +/-

Use of Medications
The mean number of medications (SD) taken was higher in the patients in the hypoglycemic group (mean=15.31 +/-12.21) as compared to those in the nonhypoglycemic group (mean=10.97 +/-9.63; t=56.78; p value 0.001). We further assessed the use of diabetes and non-diabetes medications (Tables 3.5 -3.7). Overall, the use of diabetic medications was common. Moreover, the use of these medications was significantly higher in the hypoglycemic group as compared to the non-hypoglycemic group. Specifically, 2,873 (9.5%) patients were using insulin with the use being significantly higher in the hypoglycemic group as compared to the non-

Results Of Multivariable Logistic Regression
The results of the multivariable logistic regression analyses are presented in Table   3.7. While using the age group of 18 -34 years as our reference, it was seen that increased relative risk for a serious hypoglycemic events.

Discussion
Our cross sectional study sought to identify the principal predictors of serious Previous literature has found several clinical, physiological as well as drug related factors to have a significant impact on the rates of hypoglycemia in patients. 190,194,195 Some of the important predictors that we identified in our study were comorbidities including congestive heart failure, peripheral vascular disorders, hypertension,  173 Similarly, renal failure has also been shown to have a significant association with incidence of hypoglycemia. 197 For example, Durán-Nah and colleagues attempted to identify risk factors associated with symptomatic hypoglycemia and found a threefold increase in the rate of hypoglycemia in patients suffering from renal failure (AOR 3.0; 95% CI 1.20 to 7.70). 198 Furthermore, our findings are also consistent with other studies, which have previously identified various macrovascular, microvascular, as well as other comorbidities as important risk factors for hypoglycemia. 166,169,199,200 Our results were in contrast with some studies that have previously found a protective effect with an increase in the Body Mass Index (BMI). 172,197,201 178,[200][201][202][203] Previous research suggests that along with the physiological changes that occur in the body with advancing age, other mechanisms that might contribute to incidence of hypoglycemia include decreased hypoglycemia awareness as well as decreased counter regulatory response to low blood glucose. 175,204 However, in contrast some studies hypothesize age to have a protective effect with the risk of hypoglycemia decreasing with increasing age. 173,198 Other studies demonstrate no association between age and hypoglycemia. 166,205 In our study, we found a clinically significant (but non-statistically significant) trend towards a protective effect. While evaluating gender as a predictor, there have been conflicting results in literature. For example, studies including the ADVANCE trial demonstrated no significant association between gender and the likelihood of hypoglycemia. 166,197,201 On the other hand, the ACCORD trial, demonstrated a higher risk of a hypoglycemic event for women (Hazard Ratio= 1.21, 95% CI 1.02 to 1.43). 200 Some observational studies have provided evidence of a lower rate of hypoglycemia in females as compared to males. 173,205 A possible explanation that is postulated for this gender variation in the incidence of hypoglycemia is that even though women are more likely to use health care resources to a higher degree, it might be a result of a higher morbidity rate compared with men. 206 On the other hand there are known physiological differences between males and females with a known relative reduction of counter regulatory responses in females that might also contribute to the incidence of hypoglycemia. 173,207 There has been considerable evidence that various treatment modalities might result in a higher propensity to suffer from hypoglycemia through various mechanisms like increase in insulin sensitivity. 208 In our study, we found that a higher number of medications used by the patient increased the risk of hypoglycemia. This finding is corroborated by other studies, which provided similar evidence regarding polypharmacy. 209,210 For example, Shorr et al in a study of 19,932 Tennessee Medicaid enrollees, aged 65 years or older found that patients using drugs from 5 or more therapeutic classes had a 30% increase in the likelihood of having a hypoglycemic event. 211 With respect to oral antidiabetic medications, we found a significantly increased risk with insulin, sulfonlyureas and meglitinides. Even though tight glycemic control Is being increasingly recommended in clinical practice, this has shown to pre-dispose patients to an increased risk of hypoglycemia. 208 For example, a prospectively planned group-level meta-analysis of various largescale trials demonstrated a two-fold increase in the risk of developing severe hypoglycemia in patients who were underwent intensive glucose lowering therapy. 212 Similarly, both prospective trials and retrospective studies have consistently demonstrated the higher risk of hypoglycemia events with insulin. 169,199,[213][214][215] This risk however varies with the patient's medication regimen as well as the severity of the person's diabetes. 190 However, fear of hypoglycemia is known to prolong the initiation of insulin therapy in patients thereby seriously compromising the achievement of glycemic goals. 216 Similarly, sulfonylureas have also traditionally been associated with an additional risk of hypoglycemia. 169,190,215,217  hypoglycemia as compared to insulin and sulfonylureas. 94 Interestingly, we found no significant association between the biguanide class of medications and incidence of hypoglycemia. There has been prior evidence of these drugs having a reduced hypoglycemic effect and hence are usually used as first line therapy. We did not find any association between the use of non-diabetic medications like allopurinol, warfarin, fibrates, NSAIDs, or B-blockers with severe hypoglycemia in multivariate analyses. The use drugs have previously shown an increase the likelihood of incidence of hypoglycemia. 166,190,218 Moreover, incidence of hypoglycemia has been associated with significant health care resource use as well as direct and indirect economic burden in previous literature. 184,208,219,220 Due to under reporting of hypoglycemia itself, these cost estimates are often underestimated. In our study, hypoglycemia related visits were accountable for 0.7% and 0.2% of the total costs for inpatient and outpatient visits respectively. Our results might differ from results described in other studies due to variations in the methods to measure costs as well as definitions of hypoglycemia.
For example, we calculated the cost estimates based on total amount that was paid for the services, which included the copay amounts. Heaton 184 Similarly, in a series of studies conducted in three European countries, the average cost per hypoglycemic event was found to be €537-688. 222,223 It was interesting to note that most of the hypoglycemia cost studies have been performed in patients who are being treated with insulin with there being a scarcity of information on patients who are being treated with non-insulin therapies.
Even though, the study provides valuable insights into the predictors and costs of hypoglycemia in inpatient and outpatient settings, there are some inherent limitations to our study. Since our data represents a regional health plan, the results might not be generalizable to all patients with T2DM. Secondly, as with other studies, our diagnoses of hypoglycemia only represents the most severe cases for which medical assistance was necessary since only these may be considered as reliable events. 174,190,224 The true rates of hypoglycemia may be considerably higher than our estimates. Moreover, we could not elucidate the effects of certain clinical as demographic aspects like blood glucose levels and race/ethnicity that are vital in the progression of diabetes. Lastly, due to the cross sectional design of the study, consideration of previous hypoglycemic events was beyond the scope of the study.

Conclusions
Hypoglycemia is the principal and often underreported limiting factor in the management of patients with T2DM. Considering the clinical and economic implications of hypoglycemia, we conducted a cross sectional study to identify the predictors and estimate the costs of a comprehensive definition of severe hypoglycemia. Our study confirms that specific comorbidities as well as diabetic and non-diabetic treatment modalities are significantly predictive of hypoglycemic episodes. The cost estimates also provide evidence of the significant economic burden associated with hypoglycemia. The inpatient episodes related to hypoglycemia incur a much larger financial burden as compared to the outpatient episodes of hypoglycemia. Considering the clinical burden of hypoglycemia, reducing the incidence of this adverse event in diabetic patients will have a significant impact on improvement of the quality of life of patients.           a: Cost category identified using ICD-9 codes associated with the claim and creating 3 mutually exclusive groups: 1) those identified as hypoglycemia (hypoglycemia costs).
3) others (all other costs). b: Total costs rounded to nearest dollar. c: Costs of hypoglycemic events, other diabetes-related events, and all other events may not sum to total costs due to rounding.

Manuscript 4 Evaluating The Effects Of Hypoglycemic Episodes On The Persistence To Oral
Sulfonylurea Therapy In T2DM

Sulfonylureas
Sulfonylureas (SUs) have been the widely prescribed medications in the management of T2DM since the 1950s, with as many as 75 to 80% of patients initiated on one of these agents. 225,226 First-generation agents such as acetohexamide, chlorpropamide, and tolbutamide were very popular in the 1960s. 227  to diet alone. Subsequently, the safer second generation of SUs (e.g., glyburide, glipizide, and glimepiride) emerged in the next few decades and has largely replaced the first generation of SUs. Even though they differ from the first generation of SUs with respect to their chemical composition, both groups were found to be equivalent in their hypoglycemic effect. 90

Efficacy And Indications
Many retrospective and prospective have focused on the efficacy of SUs. Although potency of effects can vary among patients, when used as a monotherapy in patients who cannot achieve the glycemic goals by nonpharmacologic interventions, SUs tend to lower A1C by 1.5 to 2.0 percentage points and fasting plasma glucose by 60 -70 mg/dL, when used as monotherapy at maximal doses are used. 90,228,229 SUs can be used as the first line oral antidiabetic agents of choice in patients who fail to achieve adequate glycemic control using nonpharmacological measures or may also be added to a patients regimen if metformin monotherapy is contraindicated, not tolerated or does not achieve the target A1C at 3-6 months of treatment. SUs can be combined with other classes of OADs, excluding insulin secretagogues and combining daytime SUs with bedtime insulin, an increasingly popular practice lacking scientific evidence of potential advantages over insulin monotherapy, can help reduce can reduce insulin doses by half. SUs are preferred in patients who are not obese or overweight since weight gain is a major concern with this class of agents. 227 Similarly these drugs are used conservatively in the geriatric population as well as patients with impaired renal and hepatic functions and belong to pregnancy category B or C. However, according to the UKPDS, despite achieving the target achieved a A1C of <7 % in the first 3 years, only 34 % of patients attained a HbA1c <7 % at 6 years, with this number further declining to 24 % at 9 years. 97,230

Adverse Events With Sulfonylureas
This class of drugs has been historically accepted as being well tolerated with only about 2 -5 % of patients reporting primary toxicities associated with it being hypoglycemia, weight gain, B cell exhaustion and adverse cardiovascular outcomes. 231 Previous literature attributes initiation of SU therapy to a resulting increase in body weight, which also accompanies many of agents, apart from metformin, that are used for diabetes management. 232 Typically, various studies report a weight gain of approximately 1 -4 kg. 227,[231][232][233][234][235] There have been concerns regarding the cardiac safety of the sulfonylureas class of drugs especially after research on the physiological effect of these drugs. 99, [236][237][238] SUs usually affect an initial increase in the B cell function, which is followed by a gradual and linear reduction, which goes hand in hand with the therapeutic failure of these drugs and the resulting progressive worsening of glycemic control. 217,239,240

Hypoglycemia with Sulfonylureas
Even though different SUs possess different pharmacotherapeutic profiles, the major, reported that more than 90% of the 57 type 2 diabetic patients experiencing glibenclamide-associated hypoglycemia were older than 60 years and more than 70% were older than 70 years. 245 Bonds and colleagues, in a secondary analysis of the ACCORD clinical trial data, did not find any association between sulfonylureas and severe hypoglycemia in both the intensive  254,255 Thus, it can be seen from the literature that the incidence of hypoglycemia is widespread in T2DM patients on SU therapy and that this has the potential to be the limiting factor in the attainment of the target goals for glycemic control. Thus, experts largely advice prudent setting of glycemic targets as well as careful selection and dosing of the oral diabetic agents based on various patient and other environmental factors.

Persistence with Sulfonylureas
The relationship between persistence to SU therapy and health outcomes has been previously explored in a number of studies. Boccuzzi et al. noted that in patients who continuously refilled a prescription for their initial oral antidiabetic within 1.5 times the days' supply of the previous fill, the rate of discontinuation for sulfonylureas was less than that for metformin (11.3% vs 11.9% 258 Hertz and colleagues conducted a retrospective cohort study of 6090 newly treated patients aged 18-64 years to determine adherence with pharmacotherapy for T2DM found that 9.7% of patients initiated on sulfonylureas were non-persistent at an early stage while almost 34% of the patients were non-persistent at the end of a year. 259 Grégoire et al. assessed persistence patterns with oral antidiabetic medications in a population-based cohort study. They found that the likelihood of continuing the with the initially prescribed oral antidiabetic medication over a 12 month period was 56% for SUs, which was about 10% less than that for metformin. In another study, Ligueros-Saylan and colleagues found that almost 44% of the patients on SUs discontinued their therapy within one year of initiation. 260 The risk of medication discontinuation during the follow up period was significantly higher for patients who were prescribed sulfonylureas as compared to metformin, the likelihood of (Adjusted conducted a study in a non-managed care setting and concluded that higher degree of medication adherence was associated with reduced use of emergency department and inpatient visits. 262 Thus considering the association between medication persistence and positive health outcomes including achievement of glycemic control and reducing complications and hospitalizations as well as medical costs, it becomes necessary to explore the association between medication persistence to SUs and various patient related factors, one of them being hypoglycemic events.

Rationale And Objectives Of The Study
Data suggests that intensive glycemic control, though beneficial in minimizing several complications of diabetes may sometimes act as a precursor to incidence of hypoglycemia. 95,200,263 Recurrent hypoglycemia is associated with significant morbidity and often leads to negative health outcomes with respect to ideal diabetes

Datasource
For the purpose of our study, we used the Blue Cross and Blue Shield Of Rhode Island

Research Design And Study Population
Using the inpatient and outpatient medical claims, and prescription claims, we conducted a retrospective cohort study using a new user design (Figure 4.1). 266 As our focus was on patients taking SU therapy, we further identified patients in the dataset who initiated treatment with a SU. More specifically, we utilized the following inclusion and exclusion criteria: Inclusion Criteria: • At least 18 years of age. Patient age was calculated as of the date of first SU fill.
• At least 1 prescription claim for an SU Oral Antidiabetic Drug (OAD).
• Continuous enrollment in the health plan for at least 12 months after the initial SU fill date.
Since we required a continous eliglibility period of 12 months following the cohort entry for each patient, we constructed eligibility episodes for each unique patient by considering the eligibility episode separate if there was a lag of 30 or more days between the end date of the previous episode and the start date of the next episode.
We only considered the eligibility episode in which the first prescription was recorded. Thus, after the initial identification of the patient cohort, we created the cohort entry date as an indicatior of the first date of fill for a SU. We further divided the study period for each patient into two phases: the baseline period and the post baseline period (Figure 4.2). The duration of the baseline phase was 3 months from the date of the first SU fill. The length of the post baseline phase was 9 months from the end date of baseline arm of the study. Thus, we followed each identified patient for a total maximum duration of 12 months starting on the first date of SU fill. An overview of study design is presented in Figure 4.2. Further, we assessed the hypoglycemia exposure, demogrpahics, other diabetic medication exposure as well as history of comorbid conditions within the baseline arm or the first 3 months of the study after the first date of fill.

Exposure To The Medication Class Of Interest (SUs)
As mentioned earlier, the BCBSRI data includes information on the use of prescription drugs by the patients in the health plan and can be linked to the inpatient and outpatients medical claim files using a unque indentifier. Utilizing the prescription claims data, we identified the use of medications in the patient population by using a combination of NDC codes as well as generic and brand product names. NDC codes are unique, three-segment numbers, usually 10 digits in length, which act as a universal product identifier for drugs. Further, for verification purposes, we used the Redbook 2008 to link the therapeutic groups and therapeutic classes to the drug records in the dataset. Moreover, we excluded observations if patient had day of supply less than or equal to 0 as well as those patients who had days of supply more than 180. Using the aforementioned methods, we identified all the patients who received their first prescription for SU along the duration of the study. It has been noted in literature that patients with T2DM are often treated with a variety of concomittant medication. Based on previous evidence, we established and outlined certain classes of drugs that were know to act as contributing factors to risk of hypoglycemia.
Once we identified the eligible study population, we then evaluated exposure history during the baseline period to create two exposure groups: 1) those with at least one inpatient or outpatient claim for hypoglycemia during the baseline period; and 2) those without any claims for hypoglycemia during the baseline period. Hypoglycemic exposure was defined as the first hypoglycemic event within the baseline period requiring medical attention in either the inpatient or outpatient setting. To identify these hypoglycemic events we used the algorithm suggested by Ginde and colleagues. 187 This algorithm identifies an hypoglycemic event if any hospital and clinic visits are indicative of hypoglycemia by using standard 251.1,251.2,270.3,962.3).Further if there is absence of hypoglycmia but presence of diabetes with other specified manifestations (ICD-9 : 250.8) without other contributing diagnoses 272.7,681.XX,682.XX,686.9X,709.3,, then such incidents are also termed as a hypolgycemic event. This algorithm is well validated and demonstrated an 89% positive predictive value (PPV) (95%CI,86-92) in accurately identifying hypoglycemia visits and an exhibited an estimated 97% sensitivity and > 99 specificity. 187,263 It has been well documented that severe episodes of hypoglycemia are often followed in quick succession by further hypoglycemic events. Hence, while characterizing the subsequent events of hypoglycmia following the first incident, we counted the events that occurred with a minimum gap of 7 days between them as separate event, else the entire hypoglycemic episode was considered as a single event. Using this information, we created two groups of indivuduals, those with at least one hypoglycemic event during the baseline period and those without. As our interest was in incident discontinuation following hypoglycemia, we excluded patients with treatment discontinuation (n= 16) before an event of hypoglycemia within the baseline period.

Outcome of interest : SU Treatment Discontinuation
We followed all persons in both exposure groups for SU medication discontinuation until the end of the follow-up period. We defined discontinuation as a gap of > 30 days in SU prescription availability that occurred between consecutive prescriptions.
Usually, the gap period allowed in various retrospective studies varies from 30 to 90 days or 1.5x last days supply. 267 Based on the clinical relevence to hypoglycemia and the day's supply limts, we chose a window of 30 days to be appropriate. This period was considered as a period where the patients would not anticipate suboptimal or negative health outcomes. 267,268 If another refill of the SU was filled within the specified window from the end date of the preceeding prescription's days of supply, the patient was considered persistent. The persistence in terms of days to discontinuation was calculated for the entire SU drug class, allowing for switches within the same class (i.e. patients switched from one SU agent to another). We followed all patients to the first of three endpoints: 1) time to discontinuation; 2) a subsequent hypoglycemic episode; or 3) the end of the study period. An overview of outcome identification is presented in Figure 2.

Demographics and Comorbidities
We compared the demographic as well as clinical characteristics between the hypoglycemia exposed and the hypoglycemia non exposed cohorts during the initial baseline period of the study. This process was carried out to characterize the potential confouding effects of certain patient related factors of interest like age and gender on the the risk of discontinuing the SU therapy. Since age was recorded as the age at the end of the enrollement period, we re-calculated the age at the date of the first fill of SU for all the patients in the study.
We used the inpatient as well as outpatient claims records for characterising rhe presence of comorbidities in patients. We identified the ICD-9 codes in these files to ascertain the presence of certain macro-and microvascular complications of diabetes as well as other conditions like Addison's disease and hypothyroidism which are associated with diabetes by employing methodologies that have been applied in other studies. We calculated a overall point estimate of comorbidity by application of the Deyo adaptation of the Charlson's Comorbidity Index. 263,269 As a part of this comorbidity index, scores from a minimum of 1 to a maximum of 6 were assigned with different weights, to each of the 19 selected medical conditions based on their adjusted relative risks and were then summed into a composite score for each individual patient. In addition to the overall comorbidity score, we also evaluated the individual comorbidities that might have an association with hypoglycemia. This was done by using the Quan H. 's enhanced ICD-9 codes of the Deyo's adaptation of the Elixhauser Comorbidity Index. The performance and validity of the Charlson and Elixhauser comorbdity indices in predicting health outcomes has been previously evaluated in a vareity of population studies where they have been consistent as prognostic measures of outcomes. 270 The presence of each comorbidity were recorded as a categorical dichotomous variable.
Similar to the SU identification, we also characterized the use of other medications in the baseline arm of the study by using a combination of NDC codes, product names as well the threpautic classes and groups. The specific classes of drugs that we were interested in evaluating were angiotensin converting enzyme (ACE) inhibitors, angiotensin 2 receptor blockers (ARB), allopurinol, benzodiazepines, beta-blockers, fibrates, fluoroquinolones, non-steroidal anti-inflammatory drugs (NSAIDs), trimethoprim and warfarin. 190,208 The presence of each drug class were recorded as a categorical dichotomous variable. We also created a continous variable to identify the total number of unique medications that were used by each patient since it is a known factor affecting medication persistence.

Statistical Analyses
In our initial phase of analyses, we compared demographic and clinical characteristics between the two exposure groups to determine group comparability. Differences between exposure and non exposure group were assessed using the Pearson's chisquare test or Fisher exact test in case of the categorical variables and Student t-test or Mann Whitney U test for the continous variables. We also conducted a descriptive analysis of the outcome variable and computed the total time to discontinuation SU therapy by hypoglycemia exposure status thereby reporting the mean time to discontinuation in both groups.
Since there would be a certain lag between the first prescription of SU to the first instance of hypoglycemia in the hypoglycemia exposed cohort, a certain degree of immortal time bias was introduced due to the fact that hypoglycemia exposed cohort would not experience the discontinuation outcome prior to exposure (Appendix 4.12). To account for this phenomena, we adjusted for during immortal time in the exposed group during our statistical analysis. [271][272][273] The schematic representation of immortal time bias is shown in Figure 3. We adjusted for this immortal time bias in our statistical analyses by moving the start of follow-up or exposed time to the end of the immortal period while accounting for the time between cohort entry and exposure date as unexposed time. date of the cohort entry until the date of the hypoglycemic event and thereafter exposed until the end of follow-up. 271,273 Other confoudning variables that were found significantly associated with hypoglycemic were eligible for entry into the model if they had atleast a 10% difference between the hypoglycemia exposed and hypoglycemia unexposed groups during bivariate anlyses. We then fitted sequential models in a nested manner by means of a manual non-computer generated forward stepwise approach. We Once the preliminary model was fitted, we assessed the mulitcollinearity in the retained variables using variance inflation factor (VIF) and the Condition Index (CX).
If the value of VIF exceeded 5, we considered that there was a high degree of multicollinearity between two covariates and warranted one to be removed from the model based on the p-value and clinical importance. We further evaluated two-way interactions between biologically and clinically plausible variables that were retained in the preliminary model. The interaction terms were sequentially added and the We also performed a power and sample size calculation based on the rate of medication discontinuation. We considered two-sided alpha level constant at 0.05 and based on our analyses, set the rate of medication discontinuation at 0.60. We considered a that a 5% change in the rate of medication discontinuations between groups is a clinically meaningful difference thereby setting the regression coefficient at 0.05. Since there are no sample size estimations available that take into account for the time-varying nature of the exposure for a Cox proportional-hazards model, we estimated the power of our study based on time fixed methods.
All statistical tests were conducted with a two-tailed alpha level of 0.05, and all statistical analyses in this study were performed using SAS software (SAS Institute Inc., Cary, NC, Version 9.3). Sample size calculations were conducted using using PASS software (2014 version; NCSS, Kaysville,UT). This study was reviewed and approved as exempt by the University of Rhode Island's Institutional Review Board.

Cohort Identififcation
The study sample was comprised of 36,954 individuals. Figures 4.3   Age was comparable in the two exposure groups with a mean age (SD) at the index date of first SU prescription of 64.9 years [11.21] and 64.5 [12.48] in the hypoglycemia group and unexposed group, respectively. Gender distribution was similar across the two cohorts with 56.4% of the hypoglycemia-exposed cohort (31 patients) and 57.2% (3,363 patients) of the patients in the exposed and unexposed cohorts respectively being female (Χ 2 =0.01; p-value: 0.90). Nine patients (16.4%) in the exposed cohort were using insulin in the baseline period in contrast to the 533 patients in the unexposed cohort (9.1%; Χ 2 =3.49; p-value: 0.06).
The prevalence of the comorbidities stratified by hypoglycemic exposure is summarized in  Table 3 and were consistently higher in the hypoglycemia group versus the nonexposed group.
The overview of the microvascular and macrovascular complications associated with diabetes is displayed in  Table 4.5 outlines a brief overview of the other medications that were concomitantly used during the baseline period. Of those who exposed an event of hypoglycemia, the average number of unique medications used was (8.63 +/-3.84) which was higher than drug use by the patients who did not experience an event of hypoglycemia (7.16 +/-3.67) during the baseline period. Notably, the use of benzodiazepines were higher in the hypoglycemia group (14.6%) than in the unexposed group (9.0%; Χ2=2.03; p value=0.1535). With respect to other medications that were considered for analyses, the unexposed group was found to be using these medications to a higher degree. For example, with in case of warfarin, 5.9% in the unexposed group and 3.6% in the hypoglycemia group were using this medication (Χ2=0.51; p-value=0.47).

Discussion
Our current study sought to determine the impact of severe hypoglycemic episodes requiring medical intervention on the persistence of the oral SU therapy. According to our findings, though the patients who experience an event of hypoglycemia in the baseline period are approximately 30% more likely to discontinue their medication, this association was not statistically significant. To our knowledge, there have been no previous studies that have specifically evaluated the impact of hypoglycemia on the time to discontinuation of SU medications following hypoglycemic exposure using a new user design.
We utilized an incident user design while conducting this study. This design was preferred since it enabled us to capture all the hypoglycemia events that occurred soon after initiation of the SU therapy. Ray et al. while reviewing the new user designs, clearly state that even with medications, rate of outcomes, both beneficial as well as adverse, varies with time since the initiation of therapy with the probability of the outcome being the maximum immediately after the start of the medication regimen. 266 Hence using this design ensured that we accounted for even those patients that were more susceptible to hypoglycemic effects of this medication regimen as opposed to other designs that might have been incapable of selecting these patients. 275 For example, in a study to assess the risk of venous thromboembolism (VTE), associated with newer oral contraceptives, Suissa et al. did not distinguish between new users, repeaters and switchers and thereby found an excess risk of VTE associated with the use of third generation oral contraceptives. The authors recognized the bias in this study and attributed it to the non-differentiation between the three different user groups. 276 In addition, this type of study design also helps alleviate the need to methodically adjust for covariates that lie in the casual pathway. 266,277 Though the risk of hypoglycemia is widely recognized in the treatment of Type 1 Diabetes, there is increasing recognition of this issue in T2DM. Seaquist and Associates reported that many trials like the ACCORD and ADVANCE have suggested that T2DM patients might be at a greater risk of adverse events associated with instances of hypoglycemia, including mortality. 278 Mitchell et al. in an online survey of 1,329 T2D patients in United Kingdom (UK), found that 23% of patients who used oral glucose lowering medications in absence of insulin experienced hypoglycemic events. 279 Leckie and colleagues conducted a prospective 12-month survey of 243 employed people to examine the frequency and consequences of hypoglycemia.
They found the rate of severe hypoglycemic events to be 0.14 episodes per person per year and concluded that this rate was lower than the rates of 1.  253 Bodmer et al. in a case control study to compare the risk of lactic acidosis and hypoglycemia among patients with type 2 diabetes using oral antidiabetic drugs found the rate of mild/moderate or severe hypoglycemia was 60 per 100,000 person years for sulfonylureas. 197 Thus, in the context of the pertinent issue, the rates that were found in our study are clinical significant findings considering the number of patients that receive SU, either alone or as part of combination therapy every year even though previous literature has documented a relatively low risk of severe hypoglycemia in patients with T2DM. In addition, our results could not be satisfactorily compared to these previous studies owing to the differences in make-up of the study population, study designs, and definitions of the hypoglycemic events.
We observed the persistence rates to oral SU medications from the cohort entry date to the end of follow up period, which was a time span of 1 year. We observed that the rate of medication discontinuation in our study was about 34.0%, with almost one third of the patients in the exposed cohort discontinuing their medication within the 9 month follow-up period. The principal finding of our study indicates that the after adjusting for all relevant covariates, patients who are exposed to hypoglycemic episodes are approximately 30% more likely to discontinue their medication regimen than those who do not experience hypoglycemia, although this finding does not attain statistical significance. Rathmann and colleagues, in a study to investigate therapy persistence and other factors in DPP-4 inhibitors and SUs, found the rate of discontinuation with the SUs to be 49%, which was 10% higher than that for DPP-4 inhibitors. 281 Moreover, it should be seen that this study did not take into account the validity of the type of diabetes the patients were suffering from as well as the prescribed daily dosages of the medications 281  Similarly, it is vital to understand that simply refilling the drug doesn't necessarily indicate persistence and that there are very few suitable means to evaluate the actual medication taking behavior of people. Further, in studies of medication compliance, it can be seen that the follow up times as well as the defined gap in medication refill in terms number of days can vary largely. In contrast to other studies, we regarded a switch to other oral medications as a discontinuation of SU therapy, potentially accounting for variation in rates between our study and previously published studies.
While our study was intended to principally analyze the impact of hypoglycemia on medication discontinuation, we found some other factors that were significantly associated with discontinuation of medication namely age, number of concomitant medications as well as use of insulin in the baseline period. It could be seen that the increase in age was significantly associated with better persistence to medications.
This is in agreement with previous studies that have investigated the relationship of age and persistence. For instance, Guénette et al. reported that patients aged 54 years or above were more likely to be persistent to their antidiabetic medications as compared with those aged from 18 to 53 with people over the age of 75 being more than 44% more likely to persist with their meducations. 110 Similarly, Hertz et al.
demonstrated that younger age was significantly associated with discontinuation of medication with patients between ages 17 -24 being much more likely to have discontinued therapy (HR=2.44, 95% CI 1.89 -3.15) as compared to patients between ages 50 and 64. 259 One possible explanation for this finding would be greater realization of the dependency of their health on therapy as compared to people who are younger. Our finding that treatment with insulin was associated with higher likelihood of medication discontinuation is supported by previous studies. Catalan et al. reported that elderly patients with previous insulin use were 1.59 times less likely to persist with acarbose drug regimen as compared to those who had not received insulin. 282 Moreover, addition of insulin to SU might exacerbate the risk of having a hypoglycemic event thereby increasing the risk of discontinuation. In context of the progression of diabetes, poly-pharmacy to meet treatment goals is unavoidable.
Hence, an interesting finding in our study was an increase in likelihood of continuing with therapy when the number of concomitant medications increased. There have been contrasting results with respect to polypharmacy and hence treatment complexity potentially being a cause of decreased therapy persistence with medications. For example, Dailey et al., Donnan et al., and Venturini et al. reported decreased compliance with increase in the number of medications. 256,283,284 On the other hand, Guénette et al. and Hertz et al. reported an increase in the persistence rate with higher number of concomitant medications. 110, 259 Grant el at. found no association between multiple medications and sub optimal treatment adherence but suggested rather that side effects of certain drugs caused the non-adherent behavior. 285 There is a plethora of literature that provides evidence that complex multifactorial relationship between achievement of glycemic objectives, medication compliance, and adverse events like hypoglycemia that might in turn affect them. There might be a range of clinical and behavioral aspects like adverse events, severity of illness and frequency of dosing that might be responsible for a person to adhere to the medication regimen. Moreover, with the oral antidiabetic medications, hypoglycemia, and fear of hypoglycemia are known limiting factors affecting the rates of medication compliance. Medication tolerability issues like hypoglycemia might in turn affect an individual's perception to perform certain self-care behaviors that are essential in the management of diabetes, adhering to medications being one of them 265 . Certain studies have demonstrated that the additional burden placed on the patients as a consequence of the physical and psychological distress that is associated with incidence of hypoglycemia and its management might prompt them to make decisions in order to balance the unpleasant effects with achievement of glycemic goals. 190 In many cases it has been observed that blood glucose levels are eventually compromised in this process. 286 Shui et al. in a cross sectional study to investigate the fear of hypoglycaemia among 120 insulin-treated patients noted that 15% of respondents reported high fear with 19.2% of the patients compromised on their blood glucose levels. 287,288 Leiter and colleagues conducted a study to assess the influence of hypoglycemia and fear of future hypoglycemic episodes on patients with type 1 or insulin-treated type 2 diabetes. The authors found that among the 133 insulin-treated type 2 diabetes patients, 29.9% and 84.2% patients reported an increased fear of future hypoglycemia following a mild and severe episode of hypoglycemia respectively. The authors also suggested that this subsequently changed the patient's willingness to continue with therapy. in a Swedish study of 309 patients above the age of 35, Lundkvist et al. found that in patients who had incidents of hypoglycemia there was lower control of diabetes, worse general health and that these patients were more anxious about future hypoglycemic events than those without hypoglycemia. The authors concluded such patients exhibited a greater degree of avoidance behavior 289 . Severity of the hypoglycemic episode often leads to reduced satisfaction to diabetes medication regimen, which in turn might result in a greater degree of medication discontinuation. 290 Although not all the aspects of the relationship between fear of hypoglycemia and compliance with the medication regimen are entirely clear, it is plausible that it acts as a major barrier to patient's medication taking behavior and thus the management of diabetes.
.However, since this was a retrospective cohort study using a medical claims database, some inherent limitations need to be taken into consideration. With respect to the diagnosis of hypoglycemia, it vital to understand that many of the hypoglycemia episodes resolve by themselves or get treated without a visit to any medical facility. Hence, our estimates of hypoglycemia might represent only a fraction of the true extent of this adverse event associated with diabetes therapy.
Secondly, due to the nature of the datasource, examination of the clinical measures of glycemic control and disease severity were not possible and hence the impact of these measures on the likelihood of medication discontinuation could not be ascertained. Moreover, since the patients might have been on SU medications while in other insurance plans, using the incident user design does not necessarily guarantee that all the patients in the final sample are newly prescribed SU therapy.
Our analysis was primarily based on the data from pharmacy drug dispensing. Hence, even though we considered failure of the patients to refill their medications as being a marker of discontinuation of therapy, in reality, there are no plausible methods to understand if this reflected the true use of medications, which might be overestimated in case of patients who filled their prescriptions but did not take their medications. Also, extending our selection of a 3 month period as a baseline period, might have added to the capability to examine a higher number of hypoglycemia events. However, it has been seen in previous studies of chronic disease medication persistence that majority of events occur within the initial period of medication initiation. In addition to this consideration, the nature of hypoglycemia episodes in patients as well as the issue of repeated hypoglycemia events was instrumental in choosing a baseline period of 3 months.

Conclusions
Hypoglycemia is a known barrier to optimal use of oral antidiabetic medications, especially SUs. Even though we found no statistical association between the incidence of hypoglycemia and subsequent persistence to SUs, it is important to take into account the implications of hypoglycemic episodes on medication adherence.
Optimal diabetes management requires timely examination of blood glucose levels and reduction in the risk of side effects like hypoglycemia when on therapy. This might be beneficial in the long-term well-being as well as the improvement in the Quality of Life (QOL) of patients. Further research must be directed towards exploring the association between hypoglycemia and other risk factors that might have an impact on medication adherence and persistence. This might provide invaluable insights into selection of an appropriate medication regimen that would provide effective glycemic control and reduction in the risk of sub optimal health outcomes.