Real-World Analysis of Antipsychotic Drugs’ Effect on Weight Gain: An EHR Application

Introduction

An estimated 57.8 million persons in the United States who were 18 years of age or older had any mental illness (AMI) in 2021¹⁸. About 20% of the US adults and 16.5% of youth suffer from severe and persistent mental illnesses, such as schizophrenia and bipolar disorder, associated with an increased risk of obesity¹⁶. Those with mental health illnesses are frequently administered antipsychotic drugs. Although the symptoms of these illnesses can sometimes be effectively treated with these drugs, they frequently have undesirable side effects such as weight gain. Hence, one of the most important reasons for weight gain in these disease conditions is Antipsychotic-Induced Weight Gain (AIWG). Administering psychotropic drugs is the leading cause of weight gain, obesity, diabetes mellitus, dyslipidemia, and metabolic syndrome ¹. The effect of Antipsychotic drugs (APD) can vary based on the demographics, such as age, ethnicity, gender, history of diabetes and hypertension, etc. The two classes of antipsychotic drugs that differ in their pharmacodynamic characteristics are first-generation antipsychotics (typical APD) and second-generation antipsychotics (atypical APD)¹. Clozapine and olanzapine (atypical APD) have the highest potential for causing metabolic side effects, including weight gain, whereas other medications, such as aripiprazole and ziprasidone, have substantially lower risks².

Some studies revealed that weight gain induces metabolic disturbances such as variations in lipid and glucose metabolism². This indicates that a misleading correlation between weight gain and antipsychotic efficacy may be caused by confounding variables such as blood lipid levels, body weight, comorbidities, and concurrent drugs³. Weight gain is likely a factor leading to drug nonadherence and discontinuation of the treatment. These factors have been associated with increased relapse and hospitalization rates, thus also resulting in higher healthcare costs^12,16.

However, aside from drug-specific clinical trials, there is limited real world evidence of AIWG leveraging large-scale observational database represented by a more generalizable population, in particular, the differential effect comparison between first-generation (typical) and second-generation (atypical) APD as well as polydrug use. In this work, we aimed to interrogate single-center Electronic Health Records (EHRs) to perform real-world analysis on the likelihood of Body Mass Index (BMI) gain after a full-year APD exposure. We compared the results between two types of risk adjustment models: a) covariate-adjusted model without propensity score weighting; b) double robust model with propensity score weighting (with reference to the univariate unadjusted model). In addition to that, the final models also shed lights on independent effects of other covariates on BMI change (such as diabetes, hypertension, drug count, and laboratory parameters like total cholesterol levels, triglycerides, HDL-C, and LDL-C).

Method

Study Cohort

The study cohort was extracted from the EHR database (Cerner Millennium) at the University of Missouri Health Care Hospital in Missouri, USA, between January 1 and December 31, 2019.This study received approval from the Missouri Ethics Institutional Review Board (IRB) and complied with all applicable international standards for ethical guidelines. The patients included in the study met the following inclusion criteria: 1) exposed to any type of antipsychotic drugs (APD) within the study period; 2) age at first APD between 18–89 years; 3) at least 2 distinct BMI records were observed before first APD prescription and at least 2 BMI records after 1-year APD exposure; and 4) at least 1 required lab test record for LDL-C, HDL-C, triglyceride, and total cholesterol at baseline.

Exposure and Outcome

We have included a list of 39 APD drugs into the study and classified them into first-generation APD (APD1G) and second-generation APD (APD2G) (Table 1). The exposure period is 1-year since they first started APD. We created indicators of APD1G and APD2G, as well as total counts of distinct generic APD1G and APD2G drugs as potential proxies for their drug switching or polydrug use behavior. The primary outcome of interests was the likelihood of BMI gain (BMI_after – BMI_before) after 1-year APD exposure. The median of all repeated BMI measurements taken during the baseline year was used to compute the value of BMIbefore, and the median of all repeated BMI measurements taken during the second year was used to calculate the value of BMIafter.

Table 1.

List of 39 antipsychotic drugs included in the study.

APD1G (first-generation APD)	APD2G (second-generation APD)
chlorpromazine	asenapine
fluphenazine	clozapine
perphenazine	iloperidone
prochlorperazine	lumateperone
thioridazine	lurasidone
thiothixene	olanzapine
trifluoperazine	paliperidone
haloperidol	quetiapine
molindone	risperidone
benperidol	aripiprazole
clopenthixol	brexpiprazole
clorprothixene	cariprazine
droperidol	amisulpride
flupenthixol	sertindole
levomepromazine	ziprasidone
mesoridazine	zotepine
clotiapine
methotrimeprazine
pericyazine
pimozide
trifluperidol
triflupromazine
zuclopenthixol

Study Design

In this retrospective, observational study, the patients were observed within three periods (Figure 1):

• –First, any time before the patient started taking APD medication (baseline period). During this period, we required at least two BMI records within 1-year prior to the start of APD medication to calculate BMI_before. During this time, all of the preselected baseline covariates were also collected.
• –The second was a year period when the patient was under an APD prescription (exposure period). This duration was dedicated to observing the effect of APD usage on BMI change. We distinguished between the APD drug classes as patients taking first-generation antipsychotic drugs only (first-generation APD), second-generation antipsychotic drugs only (second-generation APD), and patients prescribed both first and second-generation antipsychotic drugs (first-generation APD and 2). These drug classes were recorded at any time within one year after the initiation of APD. To increase the granularity of the study, specific drug counts (measured as number of distinct prescriptions) and exposure (measured as the duration of using specific drug class) were collected for both first-generation and second-generation antipsychotic drugs.
• –Third, the follow-up period after 1-years of APD exposure (post-exposure period). During this period, we required at least two BMI records to calculate the BMI_after.

Figure 1.

Study Design Timeline

Covariates

Based literature review and domain knowledge, the covariates selected for this study included: age at first APD, gender, race, ethnicity, antipsychotic drug use (indicators of first- or second-generation, total number of distinct drugs used during the 1-year exposure period), indicators of history of diabetes mellitus and hypertension, indicators of abnormal laboratory values of LDL-C, HDL-C, triglycerides, and total cholesterol levels. The abnormal lab values were defined following general guidelines with:

1. HDL-C ≥ 50-60 mg/DL for females and ≥ 40-60 mg/dL for males;

2. LDL-C >130 mg/dL;

3. triglycerides >150 mg/dL; and

4. total cholesterol ≥ 200 mg/dL.

Statistical Analysis

Comparing marginal BMI changes across patient cohorts exposed to distinct drug classes involved the application of two-sample t-tests. To comprehensively assess the genuine impact of Antipsychotic Drug (APD) usage on BMI variations, a three-fold approach was employed, employing logistic regression models:

a) an unadjusted model (unadjusted OR) – simple univariate logistic regression models regressing on each individual covariate; b) a covariate-adjusted model without controlling for any confounders (Adjusted OR-without weighting) – multiple logistic regressions incorporating all preselected covariates; c) a doubly robust estimation model with propensity score (PS) weighting (Adjusted OR- PS-weighted). The double robust estimation model was developed using propensity score weighting technique (or inverse probability of treatment weighting) assuming that the patients’ access to APD1G or APD2G could potentially be influenced by the observed covariates. We fitted two multiple logistic regression models for estimating the propensity scores of taking APD1G (PS₁) and APD2G (PS₂), and then created a composite score PS_G = (PS₁, PS₂) as a function of the two scores. We then developed the multiple logistic regression model based on the PS_G- reweighted samples. We tried multiple formulation of (PS₁, PS₂) which did not change the results much, so we only reported the results using the simplest interaction function, (PS₁, PS₂) = PS₁ • PS₂

The study data, organized in the PCORNET Common Data Model (CDM) format, was extracted from the source research EHR database using Snowflake SQL, including preprocessing stages. R version 4.2.2 facilitated all statistical analyses. A predefined significance threshold of 0.05 was employed to ascertain statistically significant associations.

Result

Patient cohort characteristics

After applying all the inclusion and exclusion criteria, a total of 7,514 eligible patient was included in the final study cohort (Figure 2). As shown in Table 2, there were 69.1% women (n = 5195) and 30.9% males (n = 2319) in the final study cohort. The age was divided into three groups- 18-39 years (n = 2383, 31.7%), 40-69 years (n = 4270, 56.8%), and 70-89 years (n = 861, 11.5%). Most of the patients belonged to the race of Whites (86.5%), followed by Black or African American (10.5%), Others (1.8%), and Asians (0.7%). The races with negligible values such as Missing or Unknown, No Information, American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, and who refused to answer, were grouped together (0.5%). A greater number of patients were non-Hispanic (n = 7296, 98.4%) as compared to Hispanic (n = 118, 1.6%). More patients took first-generation antipsychotic drugs than second-generation antipsychotic drugs, 68.1% (n = 5120) and 26.7% (n = 2003), respectively. There was a group of patients (n = 508, 6.8%) who took both first-generation as well as second-generation antipsychotic drugs.

Figure 2.

Cohort Selection Consort Diagram

Table 2.

Baseline Characteristics of the Study Cohort

Baseline Characteristics	Overall Cohort, N = 7,514
Age group (years)
18 – 39	2383 (31.7)
40 – 69	4270 (56.8)
70 – 89	861 (11.5)
Sex
Female	5195 (69.1)
Male	2319 (30.9)
Race
White	6500 (86.5)
Black or African American	791 (10.5)
Asian	53 (0.7)
Other1	138 (1.8)
Unknown	32 (0.5)
Ethnicity
Hispanic	118 (1.6)
Non-Hispanic	7396 (98.4)
History of Diabetes
Yes	1665 (22.2)
No	5849 (77.8)
History of Hypertension
Yes	3329 (44.3)
No	4185 (55.7)
Baseline HDL-C Level
Normal	3499 (46.6)
Abnormal	4015 (53.4)
Baseline LDL-C Level
Normal	5851 (77.9)
Abnormal	1663 (22.1)
Triglycerides
Normal	4903 (65.3)
Abnormal	2611 (34.7)
Total Cholesterol
Normal	5189 (69.1)
Abnormal	2325 (30.9)
APD use
Indicator of using APD1G	5120 (68.1)
Counts of distinct APD1G, mean (sd)	0.76 (0.58)
Indicator of using APD2G	2003 (26.7)
Counts of distinct APD2G, mean (sd)	0.31 (0.56)
Indicator of using both APD1G and APD2G 508	(6.8)

¹Other race category includes: American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, Refuse to answer, and other race category in the source data system.

Marginal Association between demographics, APD drug classes and BMI change

The results showed that the average BMI before the start of APD medication was 31.98 (SD = 8.8) and the average BMI after was 31.96 (SD = 8.7). The average of different first-generation drugs prescribed to the patients, i.e., drug count, was 0.76 (SD = 0.58) as compared to 0.31 (SD = 0.56) for second-generation drugs prescription. The average BMI change before and after APD initialization is 0.03 (SD = 4.19). There is a significant change in BMI after starting the APD dose.

With an average BMI change of 0.20 (SD = 4.2), second-generation (atypical) antipsychotic drugs such as clozapine, risperidone, quetiapine, olanzapine, etc., marginally caused a significant BMI change in patients (p-value = 0.03). While first-generation (typical) antipsychotic drugs such as haloperidol, loxapine, thioridazine, thiothixene, perphenazine, chlorpromazine, etc., do not have a significant association with BMI change (p-value= 0.85), nor does the group of patients taking both first-generation and second-generation drugs (p-value= 0.47). As all the preselected variables were included, there is a need to further adjust the model by controlling for some confounders.

Associations between APD and BMI gain

Among the three comparison models, we did not observe any significant association between APD1G and BMI gain, with both the weighted and unweighted Adjusted odds ratio (OR) estimated at 1.04 [95% CI, 0.82 - 1.32; p = 0.73], 0.98 [95% CI, 0.85 - 1.13; p = 0.76], respectively (Table 4). For the effect of APD2G, only the PS-weighted, adjusted OR showed a significant negative association at 0.81 [95% CI, 0.73 - 0.90; p < 0.001] (Table 4). When considering the interaction term between APD2G indicator and APD2G accounts, it showed an additive positive effect of polydrug drug use at 1.14 [95% CI, 1.06 - 1.22; p < 0.001]. As shown in Table 5, when the patients used exactly 2 different APD2G generic drugs, the total effect size of APD2G increased to 1.05 [95% CI, 0.98 - 1.13]; when the patient uses exactly 3 different APD2G generic drugs, the total effect size increased even to a positive effect of 1.20 [95% CI, 1.12 - 1.28].

Table 4.

Comparison of 3 Regression Models

Table 5.

Effect size of polydrug use

# of Distinct APD2G	OR [95% CI]
1 APD2G	0.92 [0.86, 0.99]
2 APD2G	1.05 [0.98, 1.13]
3 APD2G	1.20 [1.12, 1.28]
4 APD2G	1.36 [1.28, 1.46]

Other significant factors correlated with BMI gain

Besides indicating the potential causal effect of APD2G use on BMI gain, the models also revealed other interesting associations. The degree of weight gain varies among different age groups, and it showed a diminishing effect: compared to the based age group of 18 - 39 years, the association with BMI gain was 0.76 [95% CI, 0.71 - 0.80; p < 0.001] for 40 - 69 age groups and 0.57 [95% CI, 0.53 - 0.62; p < 0.001] for 70- 89 age group. The adjusted model also showed a significant less chance of weight gain for male with OR of 0.88 [95% CI, 0.83 - 0.94; p < 0.001]. Comparing to White, Black or African American had an increased chance of BMI gain with OR of 1.93 [95% CI, 0.84 - 0.97; p = 0.005], while being Asian was less likely to gain BMI with OR of 0.67 [95% CI, 0.52 - 0.87; p = 0.002]. Hispanic also showed a reduced chance of weight gain with OR of 0.7 [95% CI, 0.58 - 0.85; p < 0.001].

Other factors with significant associations were having a history of diabetes and hypertension both would significantly reduce the chance of BMI gain with an OR of 0.78 [95% CI, 0.73 - 0.82; p < 0.001], and 0.94 [95% CI, 0.89 - 0.99; p = 0.01], respectively. With an abnormal LDL-C at baseline was associated with an increased chance of BMI gain with an OR of 1.12 [95% CI, 1.04 - 1.22; p < 0.001]. Having abnormal HDL-C and abnormal triglycerides level were both estimated to reduce the chance of BMI gain with an OR of 0.89 [95% CI, 0.84 - 0.94; p < 0.001], and an OR of 0.88 [95% CI, 0.82 - 0.93; p < 0.001], respectively.

Discussion

The results of this study suggest that antipsychotic drug exposure is associated with the body mass index (BMI) of an individual, dispersed among various demographic characteristics, and APD drug classes. The multiple regression model performed on a subgroup of patients stratified by different demographic characteristics, and other covariates suggests that age, gender, race, BMI record prior to the APD exposure, history of diabetes and hypertension, abnormal LDL-C, HDL-C, triglycerides levels have a significant association with weight gain. A notable linkage between weight gain and second-generation drugs, alongside their frequency of prescription during the observation period, underscores the impact of APD initialization.

Exploring age stratifications, middle-aged patients (40-69 years) exhibited the highest propensity for weight gain, followed by individuals aged 70-89 years, while younger patients (18-39 years) displayed comparatively lower weight gain tendencies. Having a higher BMI record before the patient starts the APD medication leads to reduced chances of weight gain as the BMI change threshold after APD initiation is small.

Patients who take second-generation (atypical) antipsychotic drugs with a variation in the type of atypical drug prescribed (multiple counts) over the observed period suggest prominent weight gain. In contrast, patients consuming first-generation (typical) antipsychotic medication and switching between different typical drugs demonstrated relatively minor weight gain.

Gender disparities emerged, showing that males using APD medication experienced a remarkably less weight gain as compared to females by a factor of 0.88. Noteworthy is the demographic composition of the cohort, with 69.1% being women. There is a variation in weight gain observed among the different race groups. Black and African American patients have 90% less odds of gaining weight, the Asian population has 67% less odds of weight gain, while the Other races of patients have the highest probability of weight gain, albeit with smaller representation.

Regarding comorbidities, diabetic patients taking APD medication causing weight gain is 78% less while hypertension has 94% less chance of causing weight gain in the patients. This means that patients with diabetes will more likely have a reduced risk of weight gain. When accounting for other confounders it is found that only abnormal LDL-C levels have a positive correlation with weight gain. When comparing the abnormal triglyceride and HDL-C levels, we can deduce that both will cause an almost equally diminished weight gain. It is hard to decide which among the two has a more distinguished effect.

This study acknowledges limitations, notably the preselection of variables based on existing literature, potentially omitting influential variables. Furthermore, the study’s scope does not pinpoint the differential impact of specific first- or second-generation medications on weight gain. Instead, it provides an overarching perspective on the aggregate effects of these medication categories.

Conclusion

In conclusion, antipsychotic-induced weight gain (AIWG) is a concern that needs to be addressed timely. The longterm goal is early monitoring of AIWG and further differentiate the heterogenous treatment effect of APD on different sub-population. Our work showed promising results and sets a reproducible example on the properly use of real-world data (EHRs) and causal inference framework to examine the associations between APD, APD polydrug use and BMI gain, evaluated different effects caused by APD1G and APD2G as well as discovered other potential risk factors.

Figures & Tables

Table 3.

Marginal BMI changes stratified by APD use