Insights into prescribing patterns for antidepressants: an evidence-based analysis

Abstract

Background

Antidepressants are a primary treatment for depression, yet prescribing them poses significant challenges due to the absence of clear guidelines for selecting the most suitable option for individual patients. This study aimed to analyze prescribing patterns for antidepressants across healthcare providers, including physicians, physician assistants, nurse practitioners, and pharmacists, to better understand the complex factors influencing these patterns in the management of depression.

Methods

Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed to identify variables that explained the variation in the prescribed antidepressants, utilizing a large number of claims. Models were created to identify the prescription patterns of the 14 most common antidepressants, including amitriptyline, bupropion, citalopram, desvenlafaxine, doxepin, duloxetine, escitalopram, fluoxetine, mirtazapine, nortriptyline, paroxetine, sertraline, trazodone, and venlafaxine. The accuracy of predictions was measured through the Area under the Receiver Operating Curve (AROC).

Results

Our analysis revealed several key factors influencing prescribing patterns, including patients’ comorbidities, previous medications, age, and gender. A history of high antidepressant use (four or more prior medications) was the most common factor across antidepressants. Age influenced prescribing patterns, with mirtazapine and trazodone more frequent among older patients, while fluoxetine and sertraline were more common in younger individuals. Condition-specific factors included trazodone for insomnia, and amitriptyline or nortriptyline for headaches. Paroxetine, venlafaxine, and sertraline more often prescribed to females, while bupropion and doxepin were commonly prescribed for patients with tobacco use disorder and opioid dependence. Predictive factors per medicine ranged from 51 (doxepin) to 168 (citalopram), with cross-validated AROC scores averaging 76.3%.

Conclusions

Our findings provide valuable insights into the nuanced factors that shape evidence-based antidepressant prescribing practices, offering a foundation for more personalized, effective depression treatment. Further research is needed to validate these models in other extant databases. These findings contribute to a more comprehensive understanding of antidepressant prescribing practices and have the potential to improve patient outcomes in the management of depression.

Background

Antidepressants are a major treatment for depression, but prescribing them presents challenges for healthcare providers. One of the primary challenges is the absence of clear guidelines for selecting the most suitable antidepressant for individual patients. Antidepressants vary in their mechanisms of action [1, 2], side effect profiles [3], and effectiveness [4] for different types of depression, making it difficult to determine the optimal choice. Moreover, negative studies are not published and therefore not available to practicing clinicians [5, 6]. Diagnosing depression is also complex, with 227 possible ways to meet the symptom criteria, which can vary from changes in mood, appetite, to sleep patterns [7]. Without a precise diagnosis, selecting the most appropriate antidepressant and monitoring treatment response becomes more challenging. Other factors, such as comorbidities, drug interactions, and patient preferences, further complicate the prescribing decisions. Navigating these challenges requires a personalized approach that considers the unique needs and characteristics of each patient. Personalized prescribing is crucial in ensuring the effectiveness of antidepressant treatment and minimizing the risk of adverse effects. Central to this approach is the concept of shared decision-making, which emphasizes collaboration between healthcare providers and patients in determining the most appropriate treatment options [8].

The American Psychiatric Association (APA) [9, 10] guidelines recommend interventions for the identification, treatment and management of depression across all age groups. These guidelines provide recommendations for initial treatments and alternatives if the initial approach is unsuccessful, following a trial-and-error approach. In 2022, the Department of Veterans Affairs (VA) and the Department of Defense (DoD) published clinical guidelines for the management of Major Depressive Disorder (MDD) [11], providing 36 clear, comprehensive evidence-based recommendations and key decision points. The National Institute for Health and Care Excellence (NICE) [12] guideline provides similar recommendations for the treatment of first episodes of depression, subsequent-line treatments, relapse prevention, and managing chronic depression, psychotic depression, as well as depression with coexisting personality disorders. In 2023, the Canadian Network for Mood and Anxiety Treatments (CANMAT) [13] updated its guidelines to include a focus on personalized care for individuals with MDD, taking into consideration their needs, preferences, and treatment history.

It is not always clear whether healthcare providers follow guidelines or what cues they use in their decision making, as their prescribing behavior for antidepressants is often unclear [14, 15] and frequently not in accordance with established guidelines [16]. The reasons behind this complexity include the following: First, the heterogeneity of depression as a disorder means that what works for one patient may not work for another. Second, the complex interplay of biological, psychological, and social factors in depression makes it challenging to understand which cue played an important role in antidepressant prescriptions. Finally, the trial-and-error approach can mask what factors affected healthcare providers’ prescribing patterns. Few studies have clarified how these decisions are made [17].

Predictive models based on clinical data are powerful tools for studying medical decision-making [18,19,20,21]. These models analyze vast amounts of patient data to understand how important clinical decisions are made, including identifying factors that influence prescriptions and patients who are at a higher risk of adverse reactions to antidepressants. This information allows clinicians to adjust treatment plans accordingly, potentially revolutionizing the field by providing more personalized and effective treatment options for clinicians, patients, and their families [22]. The purpose of this paper is to utilize large claims data to explain healthcare providers’ prescribing patterns for specific antidepressants.

Methods

The cohort for this study was constructed using U.S. claims data sourced from OptumLabs, encompassing a total of 3,678,082 unique patients and 10,221,145 treatment episodes [23]. The detailed cohort selection and statistical methods can be found in our previous papers [23, 24]. The study began with 71,721,417 enrollees, covering the period from January 1, 2001, to December 31, 2018. We excluded 4,574,723 members who took antidepressants without a depression diagnosis, 2,790,721 who were not enrolled for at least one year before their first antidepressant, and 385,278 patients with less than 100 days of enrollment after their episode began. Additionally, 43,677 (< 1%) were excluded due to data anomalies such as incorrect birth years. This resulted in 3,678,082 members remaining in the cohort, reporting 10,221,145 antidepressant treatment episodes. The average follow-up period post start of the antidepressant use was 2.93 years, contributing a total of 15,096,055 person-years of data. Depression diagnoses were identified using ICD-9-CM and ICD-10-CM codes, and antidepressants were classified using HEDIS-2019-NDC codes, as detailed in our previous study [23].

The study examined impact of patients’ medical history, procedures, and medications on prescription of specific antidepressants. Separate analysis was done for the 14 most common antidepressants: amitriptyline, bupropion, citalopram, desvenlafaxine, doxepin, duloxetine, escitalopram, fluoxetine, mirtazapine, nortriptyline, paroxetine, sertraline, trazodone, and venlafaxine. Combination of antidepressants and prescriptions of less common antidepressants were included in a catch-all category titled “Other”.

The independent variables in the analysis were derived from the patients’ medical history. Each diagnosis, procedure, or medication was treated as a separate binary independent variable. This comprehensive approach included 16,811 outpatient predictors derived from patients’ diagnoses, 4,364 procedure variables, and 4,253 current medications identified by their generic drug names from pharmacy claims data in the analysis [25]. The SAFE rule was used to reduce the number of variables to more manageable size before conducting regression analysis. This procedure reduces variables that are not by themselves associated with the outcome/response variable. In high dimensional data, it is computationally important to reduce the variables of interest to a subset that are likely to be selected by the Least Absolute Shrinkage and Selection Operator (LASSO) regression. A detailed description of the procedure is provided in Tibshirani et al. [26]

LASSO regression is a type of regression analysis that helps in selecting important predictors by shrinking the coefficients of less relevant variables to zero. This method is particularly useful when dealing with a large number of variables, as it simplifies the model by keeping only the most important predictors. In this study, LASSO was used to identify a small subset of 1,842 variables that explained the variation in the prescribed medication. To make sure that findings were robust, the regression was repeated in 40 randomly drawn samples, and factors that showed in 95% of the regressions are reported as robust predictors. The coefficients in the models indicate the increase in the log odds of prescribing the antidepressant when the independent variable is present. A positive coefficient means the predictor increases the likelihood of prescribing the antidepressant, while a negative one means it decreases. The size of the coefficient shows the strength of the relationship, with larger values indicating a stronger effect. The accuracy of predictions was measured through the Area under the Receiver Operating Curve (AROC). AROC values range from 0.5 to 1.0, where 0.5 indicates no predictive ability (the model performs no better than random guessing), and 1.0 indicates perfect prediction accuracy. A model with an AROC between 0.7 and 0.8 is considered to have acceptable performance and above 0.8 reflects excellent performance.

Results

This cohort included patients from all states within the U.S. Since race data was not directly available, race was assigned based on the racial composition of the individual’s county of residence. County-based racial information was available for 99.83% of the individuals. Table 1 presents the age, gender, and race distribution of the patients in the cohort. The majority of our population falls within the 41–64 age group (44.75%), with an average age of 46.54 years. The sample is predominantly female (69.36%), White (77.24%), and covered by commercial insurance (81.66%).

Table 1 Demographic Distribution of Patients(Race was inferred from county where the patient resided)

The number of robust variables used for predicting the propensity of prescribing antidepressants ranged from 51 (doxepin) to 168 (citalopram) factors. Table 2 shows the top five factors that significantly influenced the prescription of each of the 14 antidepressants. The complete list of factors affecting the prescription of each antidepressant can be found in the Appendix.

Our data indicates that healthcare providers are more inclined to prescribe the same antidepressant to patients who have previously used it. This is the top most common factor influencing the prescription of an antidepressant, with coefficients ranging from 1.83 for escitalopram to 3.71 for doxepin. The high number of previous use of antidepressants (4+) is also a common factor for prescribing some of the antidepressants, with coefficients of 1.11 for paroxetine, 0.99 for escitalopram, and 0.97 for citalopram. Mirtazapine (coefficient 1.43) and trazodone (coefficient 0.41) were more frequently prescribed to older patients, whereas fluoxetine (coefficient 0.5) and sertraline (coefficient 0.38) were commonly prescribed to younger individuals. Trazodone was more likely to be prescribed to patients with unspecified insomnia (coefficient 0.47). Amitriptyline (coefficient 0.37) and nortriptyline (coefficient 0.56) were often prescribed to patients with headaches. Paroxetine (coefficient 0.30), and venlafaxine (coefficient 0.21) were more prescribed to female patients undergoing gynecologic examination, while sertraline was more likely prescribed to pregnancy women, specifically those coded for “Supervision of other normal pregnancy” (coefficient 0.69).For patients with substance use issues, bupropion (coefficient 0.32) was a common choice for those with tobacco use disorder, and doxepin (coefficient 0.54) were frequently prescribed for individuals with opioid type dependence.

Table 3 shows the cross-validated AROC explained by the 14 LASSO regression models, each corresponding to the propensity of prescribing one of the antidepressants. The average AROC was 76.3%, ranging from 77.2% for venlafaxine to 85.0% for amitriptyline. Figure 1 shows the AROC curves for selected antidepressants including bupropion, citalopram, escitalopram, fluoxetine and sertraline. The closer the curve is to the top left corner, the better the model is at predicting the likelihood of prescribing a specific antidepressant. In Fig. 1, fluoxetine (82.9%) outperforms bupropion (77.9%), citalopram (81.0%), escitalopram (79.0%), and sertraline (81.9%).

Table 2 The top 5 factors that affect the prescription of the 15 antidepressants (numbers indicate log odds for prescriptions)

Table 3 AROC scores for the 14 LASSO regression models

Fig. 1

The AROC curves for selected antidepressants

Discussion

The study found that patterns of prescriptions can be explained by patients’ medical histories and there were significant differences in when specific medications were prescribed. The cross-validated accuracy of the models ranged from 77.2% for venlafaxine to 85.0%% for amitriptyline. The number of factors influencing clinical prescribing behaviors was large (ranged from 51 for the doxepin to 168 for citalopram). The top five factors that affected these prescriptions were presented in the Table 2 and shows that clinicians are following distinct patterns for different types of patients.

Our study developed models to understand healthcare providers’ prescribing behaviors for antidepressants by leveraging a large dataset of claims data. We found that several factors significantly influence healthcare providers’ decisions, including the patient’s previous medication history, comorbidities, age, and gender. These findings are consistent with existing literature [17, 27,28,29,30,31] and guidelines [9,10,11, 13]. For example, age and gender emerged as fundamental demographic variables, aligning with previous research by Dell’Osso et al. [17], which highlighted age as a significant predictor and gender’s influence on antidepressant prescribing. Trazodone has proven effective in older patients, offering moderate anxiety relief and sleep aid. Its off-label use is increasing for managing behavioral and psychological symptoms of dementia, or other neurological disorders [29, 32]. Mirtazapine has gained attention not only for its antidepressant effects but also for its potential benefits in underweight patients [33]. Our findings also revealed that both trazodone use and mirtazapine use were associated with older patients. Additionally, trazodone was more frequently prescribed as a sleep aid for depressed patients with unspecified insomnia and mirtazapine was more likely to be prescribed to patients experiencing weight loss. Fluoxetine and escitalopram are the only FDA-approved antidepressants for teens with depression [28]. Our data confirms this, but also indicates that sertraline was prescribed to youth, supported by recent studies [34, 35].

In general, selective serotonin reuptake inhibitors (SSRIs) are often considered safe options for treating depression during pregnancy due to their well-tolerated nature [36]. Common SSRIs used in pregnancy include citalopram, fluoxetine, sertraline, and escitalopram. Our data also indicates that sertraline is commonly prescribed during pregnancy, as coded under Supervision of other normal pregnancy and Incidental pregnant state). A study showed that venlafaxine is efficacious in managing hot flashes among women with breast cancer [37], which is consistent with our finding, as we also observed its use among breast cancer patients. The antidepressants most associated with menstrual disorders were paroxetine, venlafaxine, sertraline and their combination with mirtazapine [30]. This may explain why we found that paroxetine and venlafaxine are associated with gynecologic examination.

Depression and substance use disorders are highly comorbid conditions [38]. Antidepressants are frequently prescribed for individuals with substance use disorders, as they can target underlying mechanisms related to drug use disorders and effectively treat concurrent depression [27]. For example, bupropion is utilized in tobacco cessation therapies [39], while doxepin is recognized for alleviating the symptoms of the opiate withdrawal syndrome [40], which aligns with our findings.

Our analysis suggests that machine learning can uncover a broad range of factors affecting antidepressant prescriptions—factors often missed by traditional research or treatment guidelines. Interestingly, our prior research on remission rates found little overlap between factors influencing remission and those guiding clinicians’ prescription choices [25]. When comparing our findings to standard guidelines, such as those from APA [10], NICE (UK) [12], or Canada [13], notable differences emerge. For instance, while our study identified prior antidepressant use and response to the latest medication as key predictors of prescription decisions, these guidelines instead broadly advocate for personalized treatment without specifying detailed prescription criteria.

Our models suggested 52 to 169 factors might affect prescription of a specific antidepressant. Given this large number of factors, it may be difficult for clinicians to recall or express what factors affected their prescriptions. Unlike studies reliant on clinician recall or self-report (such as surveys), our approach objectively identifies these influences, offering potentially valuable insights for clinical practice. Future research should focus on validating these models and exploring whether the healthcare providers’ prescription patterns are beneficial to patients.

One limitation of our study is that physician-related factors [41], such as personal preferences and clinical experience, may influence decision-making regarding antidepressant prescriptions. Patient preferences are also important in treatment decisions. These factors could not be obtained from our dataset, as we relied on claims data that lacks detailed clinical information. Therefore, our findings may not fully capture the complexities of clinician prescribing behavior.

Conclusions

We identified several key factors influencing prescribing patterns for antidepressants, such as patients’ comorbidities, previous medications, age, and gender. We also identified a large number of potential factors, which underscore the complexity of treatment decisions in managing depression and highlight the importance of personalized medicine in this field. Further research is needed to validate these models in other extant databases, which will contribute to a more comprehensive understanding of antidepressant prescribing practices and improve patient outcomes.