FACTORS ASSOCIATED WITH SUBOPTIMAL SAFETY LABORATORY OF METFORMIN THERAPY

The aim of the study is to determine whether the recommended lab monitoring for metformin is performed appropriately in the ambulatory care setting and if the patient characteristics are associated with monitoring rate. A cross-sectional study was performed using a healthcare claims database. An univariate analysis by frequency and percentage assessed the characteristics of patients in our study. Also, it measured the frequency of lab monitoring: HgbA1C, CBC or B12, SCR, and optimal, defined as receiving all three tests. Bivariate analyses determined the significance of differences between those receiving and not receiving lab testing according to patient characteristics. In a prediction model, multivariate logistic modeling with backward elimination was performed to identify significant patient characteristics predicting lab monitoring, and to obtain adjusted odd ratios. Optimal lab monitoring rate during 18 months rate during the 18 month was 32.88 percent. A predictive model included age category, cardiovascular, renal, respiratory disease, mental health disorder, number of clinic visit, and medication possession rate (MPR). Elderly patients with comorbidities were more likely to receive optimal care; more frequent clinic visits and greater rates of medication adherence were associated with receiving optimal lab monitoring for metformin. iii ACKNOWLEDGMENTS I would like to acknowledge my advisor Professor Dr. Stephen Kogut for his tremendous help, guide and encouragement. Your advice and encouragement has allowed me to grow as a pharmacoeoconomiest/pharmacoepidmiologist. I could not have finished my master degree without you. I would also like to thank my committee members, Professor Rita Marocux and Professor Fatemeh Akhlaghi, for serving as my committee members. I also want to thank you for your brilliant comments and suggestions. I would also thank Dr. Eunsun Noh, who introduced me to this field and has always encouraged me since. A special thanks to my family and friends. My parents sacrificed for my education and future. Thank you for giving me a priceless opportunity to experience and challenge myself and become who I am. Thank you, my friends for being emotionally supportive throughout the school years.


LIST OF TABLES
According to the current American Diabetic Association (ADA) guidelines for diabetes, metformin is a preferred first-line treatment for the treatment naïve patients with diabetes type II. 7 Metformin use is prevalent and its safety and effectiveness has been well demonstrated. Hypoglycemia occurs less frequently with metformin than any other oral antidiabetic medications. The common side effects are diarrhea, flatulence, cobalamine deficiency, and asthenia. Serious side effects include lactic acidosis yet this condition is very rare, at 0.03 cases per 1000 patients years. 8,9,10 Although metformin is a fairly safe drug, laboratory monitoring is recommended to avoid anemia, lactic acidosis, or other complications. Vitamin B12 level is recommended to be monitored every 2-3 years and hematologic parameters should be monitored at the baseline and annually to avoid anemia. 8,9 Also, renal function test (Serum Creatinine, Scr) before initiation and annually thereafter is recommended to monitor the risk 3 of lactic acidosis. Since a substantial amount of metformin is excreted through kidney, monitoring Scr level is prudent. Additionally, glycosylated hemoglobin (HbA1C) level testing monitors efficacy of the drug; it is also a safety measure to monitor hypoglycemia or to delay or avoid further complication of diabetes. 8,9 In practice, however, metformin is often used without safety monitoring given its reputation for safety. A retrospective study of metformin use in inpatient setting presented that among 204 hospitalized metformin users, 27% had at least one absolute contraindication to metformin. 10 The most common contraindication was elevated serum creatinine concentration in 32 patients (12%).
However, metformin was discontinued in only 8 (25%) of these patients. The disconnection of clinical guideline for metformin use and real practice was found in outpatient setting as well. A cross-sectional analysis, conducted in 10 different HMOs, reported that the absence of Scr lab testing at the initiation of metformin therapy was 25.8% (95% ). 11 Also, another cross-sectional study with chronic metformin users reported the rate of missing annual Scr testing by 29%, 26%, 25% in 1999, 2000, 2001, respectively. 12 Cell blood count testing was missing more frequently, 80%, 79%, 78% in 1999, 2000, 2001, respectively. 4 Several studies had examined how recommended monitoring is practiced in a clinical setting. However, no studies have examined metformin users specifically and variables that may be associated with suboptimal monitoring. The increasing diabetes population and lower safety awareness for metformin necessitates careful assessment of metformin safety laboratory screening. Therefore, we conducted this study to determine how the practice of laboratory monitoring for metformin reflects recommended guidelines. Also, patient characteristics that are potentially associated will be identified to highlight barriers to those safety measures. We hypothesize that lab monitoring for metformin will be suboptimal and may be associated with specific patient characteristics. For these categorical variables, chi-square tests were performed to determine statistical significance of differences in proportions, according to the optimal lab monitoring outcome variables. These categories included age group, gender, types of diabetes medication use, comorbidities, and level of healthcare utilization. Multi-colinearity between these independent variables was examined by a correlation matrix and diagnostics, while the interaction among the independent variables was explored using multivariate logistic modeling.

METHODS
Predictive models for optimal lab monitoring were built using multivariate logistic regression with a backward elimination process.
All variables were first included and statistically insignificant independent variables (P>0.05) were eliminated from the model step by step. The Hosmer-Lemshow goodness of fit test assessed the validity of the model. The significant independent variables in the 9 model were reported as an adjusted odds ratio with corresponding 95% confidence intervals.
Data analysis was performed using SAS (version 9.3).

RESULTS
A total of 7068 members were selected from 14,908 members in the claims database (see flowchart year old). 13 According to the bivariate analyses, all independent variables except type of diabetes medication use and the status of insulin use were associated with optimal lab monitoring performed. In addition, multivariate logistic regression modeling revealed that several independent variables significantly impacted the performance of optimal monitoring. Those variables were age group, comorbidities, number of clinic visit and medication adherence rate.
The older age groups 65-80 year of age and 80 over were more likely to receive optimal safety monitoring while younger groups 40-65 were less likely. Cardiovascular disease, renal disease, mental disease, respiratory disease may have brought more attention from practitioners and revealed the association with higher possibility of optimal lab monitoring performed. In particular, patients with renal disease were 50 percent more likely to receive optimal care than patient without renal disease (OR 1.559, 95%CI 1.298 and 1.872).
The group with 14 or more clinic visits was nearly twice as likely to receive optimal care as the group having 7-9 clinic visits. The patients with lower medication adherence rate than 70 were 23 percent less likely to have optimal monitoring than patients with an 80-89 percent adherence rate. High medication possession rate may represent high health awareness of patients (self-motivated) and be associated with more routine clinic visits.
Other multivariate models for Scr, CBC or B12, and HgbA1c were similarly affected by age, renal disease, and number of clinic visits. Interestingly, gender was a statistically significant variable in the models assessing testing for CBC or B12 and HgbA1C. Females were more likely to receive an anemia test, given the higher prevalence of this condition in female patients. Yet, HgbA1C test cannot be explained by different disease prevalence, and it is uncertain why females appeared to receive indicated monitoring more frequently.
Overall, the recommended lab monitoring for metformin was not optimally executed in practice. The metformin users with diabetes were more likely to receive optimal lab monitoring if they were elderly with cardiovascular, renal, respiratory or mental disease, visited the clinic more than 14 times in a year and demonstrated a high adherence rate with medication. Conversely, healthcare providers have to focus on monitoring younger patients with fewer comorbidities who do not visit the clinic as often, and having lower adherence to medication. Such patients are easiest to be missed in care because healthcare encounters are infrequent and typically focus on acute medical needs. Recently, pharmacy lab monitoring alert systems and other interventions have been explored as a means to increase monitoring toward optimizing the safety of care. 16 However, a first step is for healthcare providers to recognize that metformin lab 20 motoring is suboptimal, and that relatively healthier patients may be more likely to miss required laboratory monitoring. Furthermore, it is important that providers recognize the importance of lab monitoring as an important process to promote safe medication use.
Several limitations of this study exist. First, the study was conducted in claims database that is specific to one disease state and the study period spanned only 18 months. It is not possible to generalize our results to larger populations, yet our sample represented typical diabetes patients using metformin. Secondly, the all-or-none approach to our assessment of optimal lab monitoring performed may been overly strict considering typical medical practice.
Recommended monitoring for metformin may be overly excessive and impossible to implement. Third, the study data have particular limitations. The claims database was compiled based on the paid claims. Therefore, any diagnosis, procedure, and pharmacy data that was not recorded or paid out-of-pocket was missed. Also, the results could have been biased by patients' other comorbidities or medications that can influence prescribers to order labs. In this study, we examined interaction with comorbidities and age to account for this bias.
Some may argue that another limitation to this study is the fact 21 that metformin is considered to be very safe pharmacotherapy. For example, several literature has been published that rebukes the association of lactic acidosis with metformin use 17., 18