Prescription medications can cause weight gain and other metabolic consequences. These unwanted side effects are most important for medications used to treat chronic conditions. The cumulative risk for patients can be substantial because of the length of treatment required, which is often lifelong. Weight gain is also one of the most frequency reported reasons for nonadherence with some medication classes, leading to health consequences related to inadequate treatment. Available research on medication-induced weight gain and metabolic risk is mostly from randomized trials, with short term follow-up, with comparisons between only a few medications at a time. The few longer-term studies have not utilized causal methodologies that can account for baseline and time-varying confounding and have not been large enough to assess heterogeneity of treatment effects across the lifecourse. Patients and their clinicians need better information to help inform their choices of treatment, especially for conditions for which there is discretion over which medication to start.
Using data from a network of healthcare institutions, this study will provide comprehensive, generalizable results for 6 commonly-used medication classes – diabetes medications, antidepressants, antipsychotics, antiepileptics, antihypertensives, and contraceptives. The proposed study will utilize data on a large population of adults and children initiating these medications, accommodating important subgroup analyses. The study also will apply causal methods necessary for complex longitudinal data and will provide scalable data quality methods for assessing healthcare data that can be used in large networks.
The National Patient-Centered Clinical Research Network (PCORnet) is a large, collaborative national network. The network facilitates cross-institutional research because each participating institution organizes its clinical data from electronic health records and other sources in a Common Data Model. This structure allows for straightforward combination of data across institutions. The network provides a unique opportunity to examine comparative effects of different medication classes on weight gain and metabolic risk over time, for several reasons. PCORnet has a diverse population with wide representation across age, race/ethnicity groups, comorbidities, institutional characteristics, and geography. Data on weight is available at most clinical encounters, and medication prescribing data is captured routinely; several institutions also have medication dispensing data.
For this investigation, we will incorporate data from 19 healthcare systems embedded in 3 of the PCORnet Clinical Research Networks covering a broad geographic area and diverse population. From 2011 to 2016, these institutions had longitudinal data on more than 2.1 million children and 4.3 million adults. Prior to initiating analyses, we will carefully assess data quality using a previously validated approach. This method analyzes longitudinal trajectories of weight and height to identify the presence of implausible, carried forward, duplicate and erroneous measures, and compares prescription rates to national benchmark values. Applying this method to institutions in this study will facilitate troubleshooting and remediation prior to study analyses and will allow us to refine an approach for data quality evaluation that can be scaled up to the entire PCORnet network and others like it.
Because we will analyze the 6 medication classes separately, the study will be 6 independent comparative assessments of effects between subclasses (e.g., SSRI vs. TCA) and individual medications (e.g., fluoxetine v. sertraline) within each class. We will not compare across classes; however, we will account for use of medications across classes, and we will explore joint effects of medications that affect weight and metabolic risk. We will separately assess associations in children 5 to 19 years and adults 20+ years. Using approaches that incorporate novel causal methods, we will follow patients for up to 10.5 years after initiation of medications to examine the association of initiation and more sustained use of medications with weight outcomes (primary) and diabetes incidence and change in cholesterol (secondary).
Our hypotheses are that we will successfully execute the data quality assessment, providing important information for institutions to remedy any data issues and that this method will be scalable across PCORnet. We will identify important differences in weight gain and metabolic risk across subclasses and individual medications. We anticipate that this will be the most comprehensive study of medication-induced weight gain and metabolic risk conducted to date. At the end of this study, we will develop a guide for patients and clinicians that provides precise information on weight gain and metabolic risk for each of the medication classes, comparing subclasses and commonly-prescribed individual medications across the lifecourse.