EFFECT OF HOSPITALIZATION ON THE INTEGRITY OF THE MEDICATION REGIMEN IN PATIENTS WITH DIABETES MELLITUS

Medication errors are common within the United Stat es health system. Preventable medication errors are often the result of ineffective processe s that contribute to the occurrence of adverse drug events. Care transitions, movement between settings or levels of care, present a particularly vulnerab le time for patients. Errors are frequently introduced into a patient's medication regimen during transit ions of care, including the inappropriate discontinuatio n r duplication of medications. Inappropriate discontinuation (non-persistence) of evidence based th rapies for chronic diseases places patients at an increased risk for adverse health outcomes. Previou s investigations have indicated that care transitio ns due to hospitalization have been associated with in creased rates of non-persistence, and that nonpersistent patients were at an increased risk for p o health outcomes. We conducted a matched retrospective cohort study o f patients enrolled with the commercial health insurer Blue Cross Blue Shield of Rhode Isla nd. Patients included in the study were adults at least 18 years of age with diagnosed diabetes confi rmed by outpatient medication use and a diagnosis code. We evaluated the disruptive impact of hospita lization on the medication regimen by comparing the odds of persistence with evidence based therapi es between hospitalized and non-hospitalized patients. Persistence was assessed with two medicat ion classes: angiotensin converting enzyme (ACE) inhibitors/angiotensin II receptor blockers (ARBs) and lipid lowering drugs (LLD). We classified patients with an eligible hospitalization as expose d, and matched unexposed non-hospitalized patients to the exposed cohort on the variables age, gender, Ch a lson comorbidity score and enrollment period. The primary outcomes of persistence and treatment dupli cation were assessed during the 60 day period following the hospitalized patient's discharge date . Differences in baseline characteristics and the bivariate odds of persistence were assessed between groups for the primary risk factor hospitalization as well as patient demographic and health related vari ables. We constructed multivariable logistic regression models to measure the effect of hospital iza ion on persistence with medications from each class while controlling for potential confounders a nd assessing for interaction terms. A total of 201 exposed and 199 unexposed ACE inhib tor/ARB users and 202 exposed and 199 unexposed LLD users were evaluated for persistence. Aft r adjusting for potential confounders and an interaction term between hospitalization and cardio vascular disease, hospitalization was found to be a significant risk factor for non-persistence in pati en s using ACE inhibitors/ARBs [(Beta coefficient0.931 [P = 0.0283]). Patients that were hospitalize d and had cardiovascular disease had an increased odds of persistence relative to patients that were not hospitalized and had cardiovascular disease (Od ds Ratio (OR): 2.052 [95% CI 0.384-10.972)]. Patients that were hospitalized and did not have cardiovascular disease were significantly less like ly to persist compared with patients that were not hospitalized and did not have cardiovascular diseas e (OR: 0.394 [95% CI 0.171-0.906]). The odds of persistence with LLD therapy did not differ between hospitalized patients and non-hospitalized patient s (OR: 0.961 [95% CI 0.469-1.972]). The duration of p rescription supply for study medication was found to be a confounder of the exposure and outcome rela tionship for both medication classes. Therapeutic duplication occurred infrequently with both medicat on classes regardless of exposure status and the low frequencies of duplication observed precluded l ogistic regression analysis. Our results implicate hospitalization as a risk fa ctor for non-persistence with medications treating chronic diseases in commercially insured p atients with diabetes. Interventions such as medication reconciliation that strive to improve co mmunication during transitions of care and prevent the introduction of errors into the medication regi men should continue to be implemented and evaluated.


INTRODUCTION
The conclusions heralded by The Quality of Health Care in America Committee of the Institute of Medicine in their first report "To Err is Human: Building a Safer Health System" called for a system-wide quest for improvement in the quality of healthcare in the United States. 1 In the report, the committee discerned that the majority of medical errors occur as a result of ineffective systems, processes, and conditions that lead individuals to make mistakes or fail to prevent them. In 2006, a successive report "Preventing Medication Errors" evaluated the safe, effective, and appropriate use of medications throughout a multitude of health care settings. 2 The committee estimates that on average a hospitalized patient is subject to at least one medication error per day, and that at least a quarter of all medication-related injuries are preventable. The financial burden of preventable adverse drug events (ADEs) on the United States health care system is substantial. A conservative estimate of $3.5 billion (2006 dollars) is spent annually due to in-hospital preventable ADEs. 2 A care transition is the process of shifting responsibility associated with a patient's movement between settings or level of care. 3 Approximately half of all hospital related medication errors may be attributed to inefficient communications at transitions of care. 4,5 Transitions of care jeopardize the continued accuracy of a patient's medication regimen. [6][7][8][9][10] Hospitalization places patients at risk for unintentional discontinuation of evidence based therapies for treatment or prevention of chronic diseases. 6,8 Patients undergoing an additional transition to the intensive care unit (ICU) are at a greater risk for discontinuity in chronic medication use. [6][7] The occurrence of unintended medication discrepancies at the time of hospital admission has been estimated to occur in greater than half of patients. 9,10 A prospective study by Cornish and colleagues assessed the accuracy of medication histories for all patients documented to be using at least 4 medications that were admitted from the community to a large teaching hospital in Toronto, Canada. 9 The original medication history was obtained in the emergency department by either a nurse, physician, or medical resident/student. After admission, a pharmacist, pharmacy student or medical student obtained a thorough medication history which was then compared with the original history. Discrepancies were reviewed with the admitting medical team to appropriately classify intentional and unintentional changes. Of 150 patients included in the study, 81 patients were found to have at least one discrepancy (53.6%; 95% CI 45.7%-61.6%). A total of 140 discrepancies were identified, yielding a rate of 0.93 discrepancies per patient. Of the 140 discrepancies, 8 (5.7%) were classified as severe. A similar study by Gleason et al compared pharmacist obtained medication histories after admission to histories obtained by nursing and physician staff prior to admission. 10 The proportion of patients with at least one medication discrepancy was greater in this study (69% ; 1.2 discrepancies/patient [SD: 1.5]), but discrepancies were not confirmed to be unintentional as in the study by Cornish et al. 9,10 Inappropriate alteration of the medication regimen upon admission and discharge from the hospital is associated with adverse drug events and poor health outcomes. 7,8 Boockvar et al examined the impact of care transitions on medication use in patients admitted to 2 academic hospitals from 4 different nursing homes. 8 Nursing home and hospital medical records were compared for 87 patients (122 admissions) and reviewed by 2 physicians to identify ADEs attributable to medication changes during transitions of care. A mean of 3.1 medications were altered upon transition from the nursing home to the hospital, which was greater than the 1.4 that were altered upon discharge back to the nursing home (P<0.001). Of 71 bidirectional transfers reviewed, ADEs attributable to medication changes occurred during 14 (20%) of these transfers and 7 (50%) of these medication changes were therapy discontinuations. These results suggest that patients are at risk of adverse health outcomes due to inappropriate medication discontinuity following transitions of care between institutions.
Patients admitted to a hospital have been demonstrated to incur higher rates of unintended discontinuations of medications treating chronic diseases as compared to non-hospitalized patients. 6 One population based cohort study evaluated the risk of unintentional discontinuation in patients undergoing 1 or more transitions of care. Patients were required to be continuous users for at least 1 year of at least 1 medication from 5 medication classes: statins, antiplatelets/anticoagulants, levothyroxine, respiratory inhalers, and gastric acid suppressants. Compared with non-hospitalized patients the odds of unintentional discontinuation were increased in hospitalized patients without an

National and International Focus on Medication Management During Care Transitions
The Joint Commission is an independent non-profit organization responsible for the accreditation and certification of health care organizations and programs in the United States. The Joint Commission's 2014 Hospital National Patient Safety Goal 03.06.01 specifies the maintenance and communication of accurate patient medication information. 11 Performance elements for this goal emphasize the performance of comprehensive medication reconciliation procedures. 11 The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 included legislation that provided reimbursement under the newly created Medicare Part D for medication therapy management programs. 12 A year later eleven national pharmacy professional organizations collaborated to provide a widely applicable and reimbursable definition of medication therapy management. 13 Medication reconciliation, a primary component of medication therapy management, was later defined by an expert panel representing the American Pharmacists Association and the American Society of Health System Pharmacists in 2007. 14 An abbreviated version of the joint definition states that medication reconciliation is the comprehensive evaluation of a patient's medication regimen during any change in therapy in an effort to avoid errors or interactions, as well as to observe compliance and adherence patterns. A comparison of existing and previous regimens should occur at every transition in care during which the regimen is modified. 14 The use of pharmacists or other qualified healthcare professionals for medication reconciliation purposes during care transitions presents the potential for limiting medication errors and improving health outcomes. [15][16][17][18][19][20][21][22][23][24][25][26] The

Adherence and Persistence
The International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Medication Compliance and Persistence Work Group defines compliance (synonym: adherence) as "the extent to which a patient acts in accordance with the prescribed interval and dose of a dosing regimen". 30 The ISPOR Work Group defines persistence as "the duration of time from initiation to discontinuation of therapy". Persistence analyses must also include a pre-specified limit on the number of days allowed between refills before a patient is identified as non-persistent. 30 Poor adherence to evidence based therapy has been frequently documented in outpatient populations. [31][32] Non-persistence and sub-optimal adherence prevents the full therapeutic benefit of a drug from being realized and is a cause of preventable adverse health outcomes including mortality. [33][34][35][36] Inappropriate medication taking behavior increases resource utilization and the risk of mortality, leading to parallel increases in economic costs and burdens on the health care system. 2, 37-38

Lipid Lowering and Antihypertensive Therapy in Patients with Diabetes
The prevalence of diagnosed diabetes mellitus has increased steadily from an age adjusted 2.8% of the United States non-institutionalized population in 1980 to 6.4% in 2011. 39 Persons with diabetes mellitus require appropriate lifestyle and medication interventions to mitigate an elevated risk of microvascular and macrovascular complications. 40 Management of dyslipidemia using statin therapy is recommended regardless of baseline lipid levels in patients with cardiovascular disease (CVD) or in those older than 40 years of age without CVD but that have at least one other CVD risk factor identified. 41 Statin therapy is also recommended in patients that are younger than 40 years of age having multiple CVD risk factors or having a low density lipoprotein (LDL) cholesterol level of greater than 100 mg/dL. 41 The risk of major vascular events and all cause mortality is reduced in patients with diabetes using statin therapy for either primary or secondary prevention. [42][43][44] In a meta analysis of over 18,000 patients with diabetes from 14 randomized controlled trials, statin therapy was associated with a 9% reduction in all-cause mortality and a 13% proportional reduction in vascular mortality for each millimole per liter reduction in low density lipoprotein (LDL) cholesterol. 43 Inhibitors of the renin-angiotensin system (RAS) are preferred as initial therapy for hypertension in patients with diabetes due to associated reductions in the occurrence of microvascular and macrovascular outcomes. 40 The HOPE study evaluated the use of the ACE inhibitor ramipril compared with placebo in 3,577 patients with diabetes. 45 Patients were at least 55 years of age and had a history of a prior cardiovascular event or at least one current cardiovascular disease risk factor. The study was stopped before completion due to pronounced beneficial effects in patients receiving ramipril. The combined primary outcome of myocardial infarction, stroke, or cardiovascular death was reduced by 25% (95% CI: 12-36; P=0.0004), total mortality was reduced by 24% (95% CI: 8-37%) and nephropathy was also reduced by 24% (95% CI: 3-40%; P=0.027). 45 The 2014 Standards of Diabetes Care recommend that patients with diabetes and a confirmed blood pressure greater than 140/80 mmHg have prompt initiation and titration of pharmacological antihypertensive therapy. Pharmacological therapy should include either an angiotensin-enzyme (ACE) inhibitor or an angiotensin receptor blocker (ARB), substituting one class for the other if the first is not tolerated. 40

Purpose and Hypothesis
The purpose of this study was to investigate the potential disruptive impact that hospitalization may have on medication persistence with critical medication classes used in patients that have diabetes.
In patients with confirmed use of at least one of two classes of these evidence based medications, we determined if the prescriptions were renewed in the 60 day period after discharge. Furthermore, we compared the medication discontinuation rate of hospitalized patients to that of matched patients that were not hospitalized. Additionally, we measured the rate of treatment duplication (multiple prescriptions from the same medication class) between hospitalized and non-hospitalized patients. The effect of time to follow up with a primary care physician on persistence was also examined among patients that were hospitalized.
We expected hospitalization to affect a patient's medication regimen and medication taking behavior due to multiple system related factors. Upon arrival to the hospital, a medication history is obtained by emergency room staff. The completeness and accuracy of medication histories obtained prior to or during the admission process are likely to vary depending upon systemic factors including hospital policies, procedures and staffing models. During the course of a hospitalization, medications treating chronic diseases are often suspended and new medications are added in the course of managing the acute inpatient episode. Substitution of hospital formulary medications will also occur for nonformulary drugs that a patient uses at home. As a patient is prepared for discharge, new medications added during the hospitalization may be continued and chronic medications may be resumed depending upon the patient's condition. Hospital formulary drugs should be changed back to the patient's original medication used prior to admission. Effective communication between hospital practitioners, the patient and/or caregiver, and outpatient practitioners is necessary to reduce the risk for introduction of errors into the medication regimen.
A patient's medication taking behavior, encompassing adherence and persistence with prescribed therapies, is expected to be impacted by a hospitalization. The Necessity Concerns Framework proposes that patient perceptions of their own need for treatment and the potential for adverse consequences related to treatment are the main categories of beliefs that influence patient adherence. 46 Application of this conceptual framework in research studies has shown that adherence increases with parallel increases in perceived necessity of therapy and decreases in concerns regarding the medication. 46 The experience of acute hospitalization may increase a patient's perceived disease severity, therefore increasing the likelihood of adherence after discharge. Alternatively, it is possible that a patient attributes a hospitalization to a lack of effect or adverse consequence of their medication leading to decreased adherence after discharge.
Evidence based medications for which persistence and treatment duplication were assessed in this study consisted of two classes, LLDs and ACE inhibitors/ARBs. ACE inhibitors and ARBs are used interchangeably in therapy and were regarded as a single class of medications for this study. We hypothesized that hospital admission would increase the likelihood that disruption of a patient's medication regimen would occur, thus causing unintentional discontinuation of evidence based therapies as well as duplications of drug therapy in error. We expected disruption to occur because of the many systematic modifications made to the medication regimen during hospitalization, and due to absent or sub-optimal hospital based medication reconciliation practices. We expected that along with hospitalization, a longer time to follow up with a primary care provider would be associated with a decreased likelihood of persistence due to a longer time until potential resolution of errors introduced into the medication regimen.

METHODOLOGY
The study was conducted as a matched retrospective cohort study of patients with diabetes enrolled with a commercial insurer. The data for this research were provided by Blue Cross Blue Shield of Rhode Island and contained information on enrollment and demographics, as well as pharmacy and medical claims. Patients were at least 18 years of age as of July 1, 2008 and continuously enrolled for at least 12 months between the period of July 1, 2008 to December 31, 2009. All patients were confirmed to have a diagnosis code for diabetes. International classification of diseases ninth edition  codes from the 2009 Healthcare Effectiveness Data and Information Set (HEDIS) 47 were used to identify the presence of any code indicative of diabetes or a diabetes related complication (Appendix A) throughout each patient's period of continuous enrollment. In addition, all patients were confirmed to have used a medication for the treatment of diabetes, defined as the presence of a claim for any oral or injectable hypoglycemic agent during each patient's continuous enrollment period. Patients were identified using unique identification (ID) numbers in the data file; IDs without associated values for date of birth, gender, and eligibility were removed from the study population.

Defining Exposure
The primary outcome of interest in this study was the odds of persistence with chronic medications between hospitalized and non-hospitalized patients. Medication use was evaluated prior to hospital admission with two medication classes (Appendix C), angiotensin-converting enzyme (ACE) inhibitors/angiotensin receptor blockers (ARB) and lipid lowering drugs (LLD). Inclusion into the final cohort of exposed patients required at least two prescription claims for one or more medications within one of these classes during the 180 days prior to hospitalization. Separate analyses were conducted for each medication class, allowing for patients to be included in each analysis group if medications from both classes were used during the baseline period.

Unexposed Matching
Patients without a hospital stay during the study period were unexposed and a source of potential matches for eligible exposed patients. Hospitalized patients were initially linked with all potential matches that consisted of non-hospitalized patients of the same age and gender. Of these potential matches, patients with an enrollment period that encompassed the entire 180 day baseline and 60 day post hospitalization period of the linked exposed patient were retained. Matched patients were assigned an index date identical to the relevant hospitalized patient, with entirely coincident baseline and follow up periods. At least two prescription claims for the same study medication class used by the hospitalized patient were confirmed for potential matches during the 180 day baseline period. The final matching criterion was a comorbidity score calculated using weights as described by Charlson et al and updated ICD-9 codes identified by Quan et al. 48,49 A comorbidity score was calculated for all patients and the distribution of scores was then divided into four groups. Since all patients had previously been confirmed to have a diagnosis of diabetes, the comorbidity score was calculated without diabetes diagnoses. Diabetes related complications were still included in the score calculation. The majority of patients had a minimal comorbid disease burden and the distribution of the comorbidity score was highly skewed (Appendix D). Due to the skewed distribution, the four groups were created as follows: no comorbid disease (score of 0), one comorbid disease (score of 1), patients with two comorbid diseases (score of 2), and three or more comorbid diseases (score of >3). Potential matches with the same comorbidity grouping as the hospitalized patient met all criteria and were eligible to be matched.
Matches were assigned to hospitalized patients on a one to one basis without replacement. If a hospitalized patient had multiple eligible matches, a random number was assigned to all potential matches and a final match was assigned at random.

Defining Persistence
Persistence was previously defined by the ISPOR Workgroup as "the duration of time from initiation to discontinuation of therapy". 30 In our study all patients were required to have at least two claims during the baseline period prior to the index hospitalization, and a third claim during the follow up period was indicative of continued use of the medication (persistence). Persistence was evaluated as a dichotomous variable during the 60 day period following the discharge date of the hospitalized and matched patients. Patients without a prescription claim for any medication during this period were excluded from the analysis. Persistence was confirmed if the patient filled a prescription during the 60 day follow up period for any medication within the study drug class of interest. Patients without a claim for such a prescription were classified as non-persistent.

Defining Therapeutic Duplication
Therapeutic duplication was evaluated as a dichotomous variable during the 60 day period following the discharge date of the hospitalized and matched patients. Patients without a prescription claim during the follow up period were excluded from this analysis. Therapy was considered duplicated if claims were identified for greater than one generic medication name within a study drug class (ACE inhibitors/ARBs and LLD) during the post discharge period. Changes in dosing were not captured as therapeutic duplications. To reduce the potential for misclassifying patients using ACE inhibitors/ARBs who were intentionally prescribed multiple medications within the same class, patients confirmed to have been on multiple medications within the same class during the baseline period were excluded. To reduce the potential for misclassifying patients intentionally prescribed multiple LLD, therapeutic duplication was evaluated only for statins. Statins are the most commonly used class of LLD and there is no clinical situation in which duplicating statin therapy is considered appropriate. 41

Potential Confounding Variables
Age: Parametric assessment of the relationship between the continuous variable age and the dichotomous variable persistence for patients using ACE inhibitors/ARBs depicted a non-linear relationship. Age was coded categorically into three groups for the ACE inhibitor/ARB analysis. Age was determined to have a linear relationship with persistence for patients using lipid lowering drugs, and was coded as a continuous variable.
Gender: Analyzed as a dichotomous variable.
Comorbidity Score Category: The comorbidity score was constructed based upon ICD-9 coding from the 180 day baseline period. This variable was grouped into four categories due to its skewed distribution. The comorbidity score grouping was used as a criteria for matching non-hospitalized patients to hospitalized patients.
Days' Supply: Patients in this study that received prescriptions containing a supply of medication for a period greater than the duration of follow up (60 days) were at risk for misclassification of persistence.
Hospitalized patients may have additional medication remaining from a prescription filled prior to hospitalization, and non-hospitalized patients may have filled a prescription for a duration greater than 60 days that is not due for a refill during the follow up period. Duration of prescription supply, in days, was evaluated during the 180 day baseline period and included in the analysis as a dichotomous variable to control for the potential of misclassification.

Individual Comorbid Diseases:
The presence of comorbid cardiovascular, respiratory, and mental health disease was identified during the 180 day baseline period using ICD-9 codes from the HEDIS 2009 (Appendix D). The cardiovascular disease variable comprised codes for congestive heart failure, coronary artery disease, myocardial infarction, and other forms of ischemic and non-ischemic cardiovascular disease. Respiratory disease comprised codes for bronchitis, emphysema, and asthma.
Mental health disorders included codes for schizophrenia, bipolar disorder, major depression, paranoia, psychosis, anxiety, autism, panic disorder, personality disorders, acute stress disorders, impulse control disorders, anger/aggression disorders, attention deficit disorder, and attention hyperactivity deficit disorder.

Diabetes Medication Regimen:
The outpatient diabetes medication regimen was evaluated during the 180 day baseline period and considered a surrogate for severity of disease. This variable was classified categorically into four groups: no outpatient diabetes medication use, monotherapy (no insulin use), polytherapy (no insulin use), and any insulin use.
Medication regimen complexity: Regimen complexity was calculated as the number of unique chemical entities dispensed in the 180 day timeframe preceding the index hospitalization. This variable was determined to have a non-linear relationship with persistence for both medication classes and was coded categorically into four groups based upon quartiles of distribution.

Time Until Primary Care Physician Visit:
The time until follow up with a primary care physician was assessed only for hospitalized patients during the 60 day post discharge period and presented as frequencies and percentages.

Statistical Analysis:
Statistical Analysis Software Version 9.3 was used to analyze the data. Patient characteristics of the final matched cohort were stratified by exposure (hospitalization) status and frequency distributions were presented separately for each study medication class in Table 1 (ACE inhibitors/ARBs) and Table 1a (LLD). The Pearson Chi square test was used to assess differences between groups. Continuous variables were compared using t-tests for independent samples or Satterthwaites approximate t-test for variables with unequal variance. The bivariate relationships between hospitalization and persistence, and between other potential confounders and persistence, are presented in Table 2 (ACE inhibitors/ARBs) and table 2a (LLD). Frequencies and percentages of patients persisting were presented for each characteristic as well as the bivariate odds of persistence.
Results with a P-value < 0.05 were statistically significant.
Risk factor logistic regression modeling was employed to construct two multivariable models, one each for the matched cohort of patients using ACE inhibitors/ARBs (Table 3a) and the matched cohort of patients using lipid lowering drugs (Table 3b). Hospitalization was the risk factor of interest, with all other independent variables considered potential confounders of the relationship with the dependent variable of persistence. Assessment for collinearity was conducted with all possible confounders for each model. The presence of collinearity was determined based upon identifying a large condition index (>20) with multiple variables associated with a proportion of variance > 0.50. To evaluate if inclusion of interaction terms improved the model fit, the log likelihood test was used to compare models with interaction terms to reduced models. The difference in the -2 log statistic between the full model and the reduced model was compared to the corresponding Chi square statistic with the degrees of freedom equal to the difference in the number of terms in the models. The full model contained all possible two-way interactions, and a backwards elimination process was used to remove the least significant interaction terms at each step. Only interaction terms with a p-value < 0.10 were retained in the final model. Therapeutic duplication within each medication class was evaluated and presented as frequencies and percentages. The frequencies and percentages of hospitalized patients persisting as a function of time to follow up with a primary care provider were presented separately (Table 4 and Table   4a). Time to follow up with a primary care provider was dichotomized at the median, which was the same for both hospitalized ACE inhibitor/ARB users and LLD users.

RESULTS
A total of 8,891 patients met all inclusion criteria (Figure1) and were subsequently evaluated for selection into the exposed cohort of hospitalized patients ( Figure 2). Of these patients, 270 were confirmed to have had an eligible hospitalization in conjunction with the use of medication from a study drug class prior to admission (207 ACE inhibitors/ARBs and 206 LLD). A total of 7,421 patients without a hospitalization were assessed for matching eligibility with patients in the exposed cohort. An equal number of unique non-hospitalized patients were assigned as matches for the two cohorts of hospitalized patients. After matching was performed, any exposed or unexposed patients without at least one prescription claim in the follow up period were excluded prior to analysis. The final cohort of patients using ACE inhibitors/ARBs that were evaluated for persistence consisted of 201 exposed and 199 unexposed individuals. The final cohort of patients using lipid lowering drugs that were evaluated for persistence consisted of 202 exposed and 199 unexposed individuals.

ACE Inhibitor/ARB Baseline Characteristics
The analytic cohort of ACE inhibitor/ARB users was comprised of 400 patients (Table 1). prescription supply of greater than 60 days compared with 10.45% in the exposed group. Nonhospitalized patients were significantly more likely to have a prescription supply of greater than 60 days (P=0.0147).

Lipid Lowering Drug Baseline Characteristics
The analytic cohort of lipid lowering drug users was composed of 401 patients (Table 1a). The exposed and unexposed patients did not differ on the matched variables of age, gender and comorbidity index grouping. The mean (standard deviation [SD]) age of hospitalized patients and non-hospitalized patients was 57.16 [7.42] and 57.23 [7.39], respectively (P=0.927). Males made up 66.83% of the hospitalized group of patients and 68.34% of the non-hospitalized group (P=0.747). The prevalence of respiratory and mental health disease was significantly different between groups. Patients with respiratory disease made up 14.85% of the hospitalized group and 7.04% of the non-hospitalized group (P=0.012). Patients with mental health disorders made up 17.33% of the hospitalized group and 8.04% of the non-hospitalized group (P=0.005). Cardiovascular disease was unbalanced between groups but the difference was not statistically significant. Patients with cardiovascular disease made up 26.24% of the hospitalized group and 18.59% of the non-hospitalized group (P=0.067). Significantly more nonhospitalized patients utilized single drug outpatient diabetes regimens, whereas significantly more hospitalized patients utilized regimens involving insulin. Of hospitalized patients, 6.44% were using no diabetes medication, 27.23% were using monotherapy without insulin, 36.63% were using polytherapy without insulin, and 29.70% were using insulin alone or in combination with other medications. Of nonhospitalized patients, 6.03% were using no diabetes medication (P=0.867), 38.69% were using monotherapy without insulin (P=0.145), 35.18% were using polytherapy without insulin (P=0.991), and 18.59% were using insulin alone or in combination with other medications (P=0.009). Hospitalized patients utilized a significantly larger mean number of distinct medications (10.79 [4.54]) compared with non-hospitalized patients [(8.28 [3.73]), (P=0.001)]. Significant differences between groups also existed across all four stratum of the regimen complexity variable. Finally, 15.58% of unexposed patients had a prescription supply of greater than 60 days compared with 9.90% in the exposed group, this difference was not statistically significant (P=0.088).  (Table 2). Hospitalization, the risk factor of interest, was associated with a decreased likelihood of persistence that did not attain statistical significance (0.733 [0.385 -1.398]). Patients receiving a prescription supply of greater than 60 days had a significantly decreased odds of persistence with study medication than those with supplies less than 60 days (0.195 [0.098 -0.392]). The mean (mean [standard deviation]) age of patients that persisted (57.02 [7.81]) was greater than those that did not (55.38 [7.63]). Increasing odds of persistence was observed within each age stratum, but none were significant. Gender was not associated with a change in the likelihood of persistence. Patients with comorbid respiratory disease were slightly more likely to persist (1.589 [0.469 -5.386]), whereas patients with a comorbid mental health diagnosis were slightly less likely to persist (0.701 [0.293 -1.673]). Comorbid cardiovascular disease was not associated with a change in the likelihood of persistence. The odds of persistence relative to patients taking no medication for their diabetes was increased in those using single drug therapy (3.091 [0.834 -11.454]) and in patients using any insulin (1.600 [0.458 -5.585]) but unchanged in those using multiple drug therapy. The mean number of distinct medications used during the baseline period was not different between persistent (9.718 [4.759]) and non-persistent patients (9.833 [4.509]). Similarly, there were no differences in odds of persistence between the four stratum of regimen complexity. Relative to patients with a comorbidity score of 0, those with a comorbidity score of 1 (1.539 [0.566 -4.187]) or 2 (1.957 [0.656 -5.832)] had a non-significantly increased odds of persistence. A comorbidity score of 3 or greater was not associated with a change in the likelihood of persistence.

Bivariate Odds of Persistence with Lipid Lowering Drugs
The bivariate relationship between the dependent variable, persistence with lipid lowering drugs, and all other variables are presented as odds ratios with 95% confidence intervals (Table 2a).
Hospitalization was not associated with any alteration in the likelihood of persistence (0.918 [0.491 -1.719]). Patients receiving a prescription supply of greater than 60 days had a significantly decreased odds of persistence with study medication than those with supplies less than 60 days (0.167 [0.083 -0.337]). The mean (mean [standard deviation]) age of patients that persisted (57.20 [7.25]) was equivalent to those that did not persist (57.16 [8.58]). Gender was not associated with a change in the likelihood of persistence. Patients with comorbid respiratory disease were slightly less likely to persist  (9.49 [4.40]) and non-persistent patients (9.98 [3.80]). Compared to patients using the fewest number of medications (less than 6) during the baseline period, those in each other stratum of regimen complexity (6-8, 9-12, 12< medications) were more likely to persist. Relative to patients with a comorbidity score of 0, those with a comorbidity score of 1 or 2 had a comparable likelihood of persistence. A comorbidity score of 3 or greater was associated with an increased odds of persistence (1.891 [0.698 -

Multivariable Logistic Regression Model: ACE Inhibitor/ARB Cohort
The results of a multivariable logistic regression analysis of the effect of hospitalization on persistence with ACE inhibitor/ARB therapy, adjusted for relevant confounders and interaction terms, are presented in Table 3. Collinearity was not found between any of the independent variables assessed for inclusion into this model. A single interaction term between the risk factor (hospitalization) and cardiovascular disease met criteria and was included in the final model. After adjusting for all potential confounders and the interaction between hospitalization and cardiovascular disease, the beta coefficient representing the relationship between hospitalization and persistence was significant ( Assessment for confounding involved comparison of the beta coefficient and P-value for hospitalization from the final model with the beta coefficient and P-value obtained from a reduced model absent the potential confounder of interest. A substantial change upon variable removal was indicative of confounding. Duration of prescription supply impacted a large magnitude change in the beta coefficient upon removal from the model. Prescription supply was also significantly associated with hospitalization, and as a result was identified as a confounder. Through this same process, the diabetes regimen and medication regimen complexity variables were also identified as confounders.
The inclusion of all other variables in the model was determined necessary to adjust for differences in baseline characteristics and due to clinical importance.

Multivariable Logistic Regression Model: Lipid Lowering Drug Cohort
The results of a multivariable logistic regression analysis of the effect of hospitalization on persistence with lipid lowering drug therapy, adjusted for relevant confounders and interaction terms, are presented in Table 3a. Collinearity was not found between any of the independent variables assessed for inclusion into this model. There were no interaction terms that met criteria for inclusion into the final model. The odds of persistence did not differ between hospitalized patients and non-hospitalized patients (0.961 [0.469-1.972]). Patients receiving a prescription supply of greater than 60 days were much less likely to persist (0.146 [0.068 -0.313]) relative to patients with a supply of less than or equal to 60 days. Patients utilizing medication to treat diabetes had an increased odds of persistence compared to patients not on drug therapy, regardless of whether it was monotherapy, polytherapy or any regimen containing insulin. The increased odds of persistence achieved statistical significance for patients using monotherapy (3.765 [1.064-13.324]). A consistently lower likelihood of persistence was observed in the three groups of patients with greater regimen complexity (6-8, 9-12, 12< medications) relative to those using 5 medications or less during the baseline period, but this result did not achieve statistical significance.
A confounding assessment was carried out using the process described for the ACE inhibitor/ARB model. Duration of prescription supply caused a large magnitude change in the beta coefficient upon removal from the model. The negative association between prescription supply and hospitalization approached significance (P = 0.088), as a result prescription supply was identified as a confounder. Through this same process, the diabetes regimen and medication regimen complexity variables were both identified as confounders. The inclusion of all other variables in the model was determined necessary to adjust for differences in baseline characteristics and due to clinical importance.

Therapeutic Duplication
Therapeutic duplication occurred infrequently with both medication classes regardless of exposure status. Of the 189 hospitalized patients that were only using a single ACE inhibitor or ARB during the baseline period, 3 (1.59%) duplicated therapy during the post-hospitalization period.
Similarly, of the 186 non-hospitalized patients that were using a single drug during the baseline period, 1 patient (0.54%) duplicated therapy during the follow-up period. Of the 186 hospitalized patients that had 2 or more claims for a statin during the baseline period, 1 patient (0.54%) duplicated therapy. A total of 3 (1.63%) non-hospitalized patients duplicated statin therapy of the 184 that had 2 or more claims for a statin during the baseline period.

Time to Primary Care Physician Visit
The number of patients persisting was similar between patients with a shorter (<12 days) time until a follow up visit with a primary care provider compared with patients with a longer (12< days) time until follow up for both ACE inhibitors/ARBS (P=0.537) and for LLDs (P=0.786) ( Table 4 and   Table 4a).  (11) a The median time until follow up with a primary care physician was 12 days; 19 patients did not follow up within the 60 day period and were grouped with the patients that had a time until follow up of 12< days We also investigated hospitalization as a possible risk factor for inappropriate duplication of therapy with both classes of study medications. The low frequencies of duplication observed [(ACE inhibitor/ARB:1.59% hospitalized, 0.54% non-hospitalized); (LLD: 0.54% hospitalized, 1.63% nonhospitalized)] precluded bivariate and multivariable analysis for this outcome. Therapeutic duplication was found to be infrequent in this population due to a confluence of factors. To reduce the likelihood of misclassification, duplication was only assessed for statins as a subclass of the broader class of LLD.
The use of multiple LLDs may be therapeutically indicated, whereas the use of more than 1 statin is not appropriate in any clinical situation. Similarly, patients that filled multiple ACE inhibitor/ARB prescriptions during the baseline 180 days were excluded from the duplication analysis during the follow up period. It was not possible to determine if the use of multiple drugs from this class was therapeutically appropriate during the baseline period. Other contributors to the observed low frequency of duplication may be the ease of detection for healthcare providers and the use of decision support software that would flag the prescription prior to dispensing. It is also possible that patients were duplicating therapy at home from previously dispensed prescriptions, in which case we would be unable to detect such inappropriate medication usage.
The duration of prescription supply emerged as a significant confounder of the association between hospitalization and persistence. The number of patients that received a prescription supply of greater than 60 days for a study medication during the baseline period was unevenly distributed between the hospitalized and non-hospitalized groups. In the bivariate and multivariable analyses, a supply duration of greater than 60 days increased the odds of non-persistence for ACE inhibitors/ARBs and LLDs. This phenomenon may be explained by continued use of a 90 day prescription that was filled before hospital admission, or in matched patients, the sustained use of a 90 day prescription that was not due for a refill during the follow up period. In either case, patients would be at risk for being misclassified as non-persistent despite appropriate use of the study medication. Inclusion of the days' supply variable in the final model was essential in order to adjust for the effect of supply duration, which if left unadjusted would have obscured the effect of hospitalization on non-persistence. In contrast, therapeutic duplication is more likely to have been underestimated as a result of misclassification in this study. Patients choosing to continue taking a medication prescribed prior to hospitalization, in addition to another medication within the same class prescribed upon discharge, would be inappropriately duplicating therapy. Such duplication is not detectable unless the patient refills the original prescription during the follow up period, which may not be necessary if the original supply was for greater than 60 days.
Evaluations of real-world medication usage by patients apply the terms adherence (synonym: compliance) and persistence to describe two separate constructs. 30 Other terms have been used to describe persistence (discontinuation rates, continuous adherence, persistency, time of continuous adherence), and reports have frequently stated that the endpoint under investigation was persistence when it was in fact adherence and vice versa. Medication adherence is the act of conforming to the recommendations of the provider with respect to timing, dosage, and frequency of medication taking.
Medication persistence refers to the duration of time from initiation to discontinuation of therapy. 30 In our study, the outcome of persistence was dichotomized and patients were categorized as persistent or non-persistent depending on the presence or absence of a prescription claim indicating therapy continuation during a pre-specified period of time. All patients were required to have at least two claims during the baseline period prior to the index hospitalization, and a third claim during the follow up period was indicative of continued use of the medication (persistence).
The healthy adherer effect postulates that improved clinical outcomes in adherers to drug therapy compared with non-adherers is not entirely attributable to the benefits of the medication. 50 Instead, adherence to medication is a surrogate marker for overall healthy behavior, which introduces bias if left unaccounted for in an analysis of drug effect. 51 The effects of healthy adherers are not evident in our results, but a separate construct that influences medication taking behavior may be implicated.  58-0.87). In this study prescription data was not available from private insurers and as a result only elderly patients were included. Rates of primary nonadherence following hospitalization for AMI were low, but these results are unlikely to represent younger or commercially insured populations. 53 The findings by Gregoire et al and Jackevicius et al reinforce the proposed connection between patient perceptions of medication efficacy and the necessity for compliance and persistence with therapy. [52][53] In the survey study by Gregoire et al both patient perceived absence of drug effectiveness and the occurrence of adverse effects attributed to the drug were associated with significantly lower persistence. 52 Instead of directly reported perceptions, patients in the study by Jackevicius et al were retrospectively selected based upon recent hospitalization specifically for AMI. 53 The finding of substantially higher primary adherence rates with cardiac medications follows directly from the recognition of elevated cardiac risk that would be anticipated in patients following a hospitalization for AMI. Insurance coverage, a predictor of persistence identified by Gregoire, 52 was not a factor in our study since all patients were confirmed to be enrolled with the same commercial insurer. Additional covariates that were not associated with changes in persistence including age, gender, and comorbid disease burden were evaluated and yielded concordant results with those obtained in our present research. 52 Identification of patients that received pre-discharge counseling, a predictor of persistence in the study by Jackevicius,53 was not possible with our commercial claims data source.
Long term use of evidence based, guideline recommended oral therapies for heart failure has been shown to reduce morbidity and mortality. 54 The Get With The Guidelines-Heart Failure program is an ongoing, prospective, observational data collection and quality improvement initiative that collects information on medical history, hospital care and clinical outcomes. 55 Krantz et al assessed the inpatient patterns of use and continuation of ACE inhibitors/ARBs, aldosterone antagonists, and beta blockers while also determining predictors of use. 56 Patients included for evaluation had reduced ejection fraction (EF<40%) heart failure and were admitted to hospitals participating in the program across the United States. In patients with no contraindication to therapy, the proportion of patients using ACE inhibitors/ARBs, beta blockers, and aldosterone antagonists at admission was 65.3%, 72.6% and 15.6%, respectively. The proportion of patients using ACE inhibitors/ARBs, beta blockers, and aldosterone antagonists at discharge was 92.9%, 90.1% and 26.2% respectively. Of the population of patients that were already receiving ACE inhibitors/ARBs at hospital admission, 2.6% did not persist with the medication at discharge. The strongest predictors of ACE/ARB usage at discharge were medication usage at admission (OR 7.4;, the absence of concomitant renal insufficiency (OR 2.7; 95% CI 2.1-3.4), and the absence of concomitant hypertension (OR 1.34; 95% CI 1.02-1.77). These results are limited by the lack of available persistence data for the period following hospital discharge. 56 In the study by Krantz et al patients with heart failure that were taking ACE inhibitors/ARBs upon admission into the hospital were more than 7 times as likely to be discharged on the medication compared with patients admitted that were not receiving a drug from this class. 56 Our study design included only patients that were confirmed to be using a medication from a study drug class prior to a hospital admission. We observed a proportion of patients with cardiovascular disease that did not persist with ACE inhibitor/ARB therapy of 10.26%, which was much larger than the 2.6% that did not persist in the study by Krantz et al. 56 Much of this difference can likely be explained by the definition of persistence, which was measured at discharge by Krantz and did not require confirmation of a prescription dispensing after the patient left the hospital. 56 Although persistence patterns with statins were not assessed in this population of heart failure patients, we found a comparable rate of nonpersistence with statins in patients with cardiovascular disease of 10.11%.
Quality of care measures provided prior to hospitalization have been shown to impact 30 day re-hospitalization rates in a nationally representative population of commercially insured adult patients with diabetes. 57  was limited by the lack of persistence data following discharge and the absence of a comparator group of non-hospitalized patients. 57 Bell et al recently evaluated the risk of unintentional discontinuation of medications prescribed to treat chronic disease in patients undergoing 1 or more transitions of care. 6 Similar to the study by Jackevicius, 53 this study utilized linked prescription claims, vital statistics, physician services, and hospitalization databases to identify elderly patients (age > 66) admitted to all acute care hospitals in Ontario, Canada. 6 Patients were required to be continuous users for at least 1 year of at least 1 medication from 5 medication classes: statins, antiplatelets/anticoagulants, levothyroxine, respiratory inhalers, and gastric acid suppressants. Exposure was categorized into three groups: non-hospitalized patients (unexposed), hospitalized patients (1 transition of care), and hospitalized patients that spent time in the ICU (2 transitions of care). The primary outcome of interest was the absence of a prescription renewal for a drug from within the original medication class in the 90 days following the index date (the date of discharge for hospitalized and a randomly assigned date for non-hospitalized patients in patients that discontinued statins (OR 1.07; 95% CI 1.03-1.11) or antiplatelets/anticoagulants (OR 1.10; 95% CI 1.03-1.16) within the 90 days after discharge. These results provide valuable context for the relationship between non-persistence and adverse health outcomes in a large representative population of elderly adults following a hospitalization. Risk for adverse events following discontinuation was not equal between medication classes, with the greatest risk observed in medications used for the prevention of macrovascular events. 6 Improving outcomes following a hospitalization, including reducing 30 day readmission rates, has become an incentivized priority with the implementation of the Affordable Care Act. 58 Underperforming hospitals with increased readmission rates for certain disease states relative to other similar institutions are subject to reduced reimbursement. 58 The results presented by Chen suggest that adherence to recommended processes of care for commercially insured patients with diabetes will positively influence 30 day readmission rates. 57 Bell and colleagues demonstrated that appropriate use of chronic medications in elderly patients is disrupted incrementally by 1 or more transitions of care.
Furthermore, non-persistence following a hospitalization with medications for the prevention of adverse macrovascular outcomes (statins, antiplatelets/anticoagulants) placed elderly patients at an increased likelihood for 1 year mortality and rehospitalization. 6 Our study evaluated a commercially insured population, similar to Chen, 57 but persistence after a hospitalization was measured with medication classes used for the prevention of micro and macrovascular outcomes, similar to Bell. 6 Considering these relevant findings, our results indicate that patients with diabetes and comorbid cardiovascular disease that were not hospitalized, as well as patients hospitalized without cardiovascular disease, may also be at an increased risk of adverse health outcomes due to lower rates of persistence. Further studies designed to follow persistence patterns and evaluate health outcomes are necessary in patients with diabetes.
Based upon our multivariable model, certain populations of LLD users are possibly at greater risk of unintentional discontinuation of LLD therapy. A non-significant trend was observed for decreasing comorbid disease burden associated with increasing odds of non-persistence (Table 3a).
Similarly, patients without documented use of a medication for diabetes treatment during the baseline period were the least likely to persist with LLD therapy. An opposite tendency was demonstrated with regards to regimen complexity. In agreement with our hypothesis, patients using the least number of medications during the baseline period had the greatest relative likelihood of persisting. These conflicting phenomena are presumed to be the consequence of multiple contributing factors and random variation. Healthier patients with diabetes that were not using hypoglycemic medication and had a minimal burden of comorbid disease may not have perceived LLD treatment as necessary and were consequently at an increased risk for non-persistence. In addition to polypharmacy, the regimen complexity variable may have captured a separate indicator of disease burden that was not fully reflected in the comorbidity score grouping and was associated with an increased risk for nonpersistence. Ultimately the main result of the LLD multivariable analysis was that after adjusting for all possible confounders, there was no difference in odds of persistence between non-hospitalized and hospitalized patients with diabetes (0.961 [0.469 -1.972]).
Studies evaluating relative rates of mortality and other adverse health outcomes between adherent/persistent patients with non-adherent/non-persistent patients have demonstrated that appropriate usage of statins and ACE inhibitors/ARBs is associated with better clinical outcomes. [6][7][33][34][35][36][37][38]  Recognition of the detrimental outcomes occurring secondary to inappropriate management of the medication regimen during care transitions has lead to the development and evaluation of numerous institution specific interventions. [15][16] Due to the fragmented nature of the United States health system and the logistical difficulties in organizing large multi-site trials, stakeholders have typically approached the issue individually as it directly relates to a specific practice population, professional discipline, or institution. Individualized programs targeting care transitions have involved a multitude of different healthcare providers including nurses, 17-19 physicians, 20 pharmacists, [17][18][19][21][22][23]   history have outperformed hospitals with physician or nursing based models. 28 This differential outcome from the same underlying process draws attention to potential difficulties encountered with a standardized approach. Expertise is not entirely coincident between practice disciplines and allocation of responsibility to specific healthcare practitioners will be inconsistent between institutions. The success of the SOP for medication reconciliation, if sufficiently validated, will still require crossdisciplinary collaboration and tailoring of the SOP to best function within individual practice sites.
The present study utilized a retrospective matched cohort design to ensure that hospitalized and non-hospitalized patients were similar with regards to age, gender, comorbidity burden and enrollment period. The purpose of matching in this study was to ensure that the group of nonhospitalized patients was similar to the group of hospitalized patients. Despite the matching procedure, a significantly higher percentage of hospitalized patients had diagnoses indicating respiratory and cardiovascular disease, and mental health disorders. While our multivariate analyses attempted to control for these differences, it is possible that the effect of hospitalization on medication persistence was biased by the greater overall disease burden among the hospitalized group. Due to the retrospective, non-randomized study design and the use of claims data it was not possible to adjust for all possible confounders. As a result, the potential for uncontrolled residual confounding existed due to additional variables that may have included but were not limited to socioeconomic status, healthcare service utilization, delivery of medication counseling, education level achieved, and patient perceptions regarding benefits and detriments of pharmacy care services. An additional limitation was the breaking of matches after the matching procedure had been completed that caused minor inequalities between the size of hospitalized and non-hospitalized patient populations in both study drug cohorts. This resulted from exclusion of patients that did not have a prescription claim during the follow up period. This procedure was necessary to prevent misclassification of patients no longer filling prescriptions with the insurer as non-persistent, but it may have further contributed to the unequal distribution of confounders between groups.
The primary outcome of interest, persistence, is a surrogate marker for adverse health outcomes that are expected to follow the inappropriate discontinuation of evidence based therapies for chronic disease. A comparison of readmission and/or mortality rates between persistent and nonpersistent patients would be useful to fully characterize the adverse effects of medication disruption by inpatient hospitalization. Moreover, the potential for misclassification surrounding the primary outcome must be considered. Although the medication classes selected for this study are recommended for use in broad populations of patients with diabetes, 39 it is possible that the medications were intentionally stopped by prescribers for legitimate reasons. Such causes may have included intolerable adverse effects, newly developed contraindications to therapy, or lifestyle control of the medical condition that precluded the necessity for continued drug treatment. Detection of prescription fills for persistence confirmation was also limited to claims submitted to the commercial insurer. Prescriptions that were bought without insurance or with alternative insurance coverage would not be detected and patients would be at risk for misclassification. We sought to mitigate such bias through the exclusion of patients that did not have a prescription claim for any medication during the follow up period. Our exclusion of such patients contributed to an underestimation of persistence if it is presumed that these patients filled their prescription without reimbursement from Blue Cross Blue Shield of Rhode Island. With claims data, adherence to prescription medications is determined using the surrogate marker of a prescription dispensing. It is possible that patients picked up medication but then did not proceed to take it, resulting in misclassification of baseline adherence or follow up persistence.
Our study is believed to be the first to evaluate persistence patterns following hospitalization in a commercially insured population with diabetes that was adherent to evidence based therapy prior to hospitalization. The use of a matched comparator group of patients with diabetes that were not hospitalized sought to preclude the introduction of bias and improved the interpretability of our results.
The capacity for generalization of our results is limited to a commercially insured population with diabetes using ACE inhibitors/ARBS or LLDs. Further research evaluating persistence and successive clinical outcomes in this population is required to better characterize the impact of regimen disruption secondary to hospitalization. Confirmation of our results is also warranted in nationally representative populations of elderly and commercially insured patients with diabetes. Future studies using outpatient claims data would be improved through an integrated analysis with inpatient and outpatient medical records, which would increase the specificity for identifying true non-persistence. Intentional medication discontinuation would be detectable in the medical chart, comorbid disease burden would be verifiable, and information on additional potential confounders would be available for assessment.
Ideally, medical records and claims data will be used together to evaluate the effect of a care transition intervention, such as medication reconciliation, in a randomized prospective study conducted across multiple institutions. Such a large scale and rigorous study methodology is necessary to generate widely applicable evidence of improved clinical outcomes and to justify funding and implementation of specific care transition interventions.
In conclusion, hospitalization was found to be a significant risk factor for ACE inhibitor/ARB discontinuation in commercially insured patients with diabetes without comorbid cardiovascular disease. Hospitalized patients with cardiovascular disease were more likely to persist with ACE inhibitor/ARB therapy than non-hospitalized patients with cardiovascular disease. Hospitalization was not found to disrupt continuation of LLD treatment after discharge, as persistence rates were similar to non-hospitalized patients. A prescription supply of greater than the number of days in the follow up period was identified as a strong confounder of persistence with both drug classes. Prescription supply duration should be considered in future studies measuring persistence. Further evaluation of the disruptive impact of hospitalization on appropriate medication use in patients with diabetes should focus on quantifying increased risk of adverse health outcomes with non-persistence and the effectiveness of care transition interventions on preventing unintentional medication discontinuation.  (35)  3 6.28 (13) 8.74 (18)  4 2.90 (6) 3.40 (7)  5 0.48 (1) 0.49 (1)  6 0.97 (2) 1.94 (4)  7 0.97 (2) 1.46 (3)