Impact statement
Major Depressive Disorder (MDD) imposes a major public health challenge worldwide, yet clinical management is frequently compromised by suboptimal adherence. A growing body of research suggests that when treatment options fail to align well with patient priorities, difficulties such as reduced engagement or dissatisfaction with care may arise. However, empirical evidence quantifying patient preferences for antidepressant attributes remains limited. This study contributes patient-centered evidence by quantifying how patients in China prioritize treatment benefits, side-effect profiles and out-of-pocket costs. Using a best-worst scaling approach, it moves beyond general discussions of patient-centered care by quantifying the relative importance patients assign to specific treatment features. The findings show that patients place greatest value on treatment efficacy and the avoidance of certain adverse effects, while cost and the risks of liver or kidney injury also influence preferences to a lesser extent. These results do not aim to prescribe specific treatment choices but rather to clarify how patients weigh key trade-offs when considering antidepressant options. By making patient preferences explicit and measurable, this study offers a reference for future research on preference-informed decision-making in mental health care. The findings may also help clinicians and health system stakeholders better understand patient priorities and support efforts to align mental health delivery more closely with patient needs.
Introduction
Major depressive disorder (MDD) is a prevalent condition that poses substantial public health and economic challenges. According to the China Mental Health Survey (2013–2015), MDD has a weighted lifetime prevalence of 3.4% and a 12-month prevalence of 2.1%, making it the most common mental disorder (Huang et al., Reference Huang, Wang, Wang, Liu, Yu, Yan, Yu, Kou, Xu, Lu, Wang, He, Xu, He, Li, Guo, Tian, Xu, Xu, Ma, Wang, Wang, Yan, Wang, Xiao, Zhou, Li, Tan, Zhang, Ma, Li, Ding, Geng, Jia, Shi, Wang, Zhang, Du, Du and Wu2019). While a wide range of antidepressants is available, clinical practice frequently faces obstacles such as treatment non-adherence and premature discontinuation. These issues may, in part, stem from the complex trade-offs patients navigate between therapeutic efficacy, medication-specific tolerability, and economic burden. For instance, while agents like mirtazapine, escitalopram, and venlafaxine demonstrate superior efficacy in alleviating depressive symptoms, they differ in tolerability profiles, ranging from appetite modulation to gastrointestinal distress (Cipriani et al., Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa, Leucht, Ruhe, Turner, Higgins, Egger, Takeshima, Hayasaka, Imai, Shinohara, Tajika, Ioannidis and Geddes2018; Oliva et al., Reference Oliva, Lippi, Paci, Del Fabro, Delvecchio, Brambilla, De Ronchi, Fanelli and Serretti2021). Consequently, aligning treatment selection with patient-specific tolerance levels is often proposed as a strategy to support clinical outcomes. Understanding how patients weigh these diverse attributes may help in optimizing personalized treatment planning.
Despite the availability of effective pharmacological treatments, suboptimal medication adherence remains a major clinical concern. Evidence suggests that approximately 30% to 60% of patients discontinue antidepressant therapy within the first three months (Demyttenaere, Reference Demyttenaere2003; Kripalani et al., Reference Kripalani, Yao and Haynes2007; Rossom et al., Reference Rossom, Shortreed, Coleman, Beck, Waitzfelder, Stewart, Ahmedani, Zeber and Simon2016), often prematurely failing to achieve sufficient therapeutic exposure. Such non-adherence frequently precipitates persistent symptoms and elevates relapse risk, thereby compromising overall clinical outcomes (Solomon et al., Reference Solomon, Keller, Leon, Mueller, Lavori, Shea, Coryell, Warshaw, Turvey, Maser and Endicott2000). Emerging evidence highlights that incorporating patient preferences into treatment decision-making is essential to fostering better adherence and therapeutic engagement (Losi et al., Reference Losi, Berra, Fornengo, Pitocco, Biricolti and Federici2021; Marasine and Sankhi, Reference Marasine and Sankhi2021; Solmi et al., Reference Solmi, Miola, Croatto, Pigato, Favaro, Fornaro, Berk, Smith, Quevedo, Maes, Correll and Carvalho2021).
Active patient involvement and the alignment of pharmacotherapy with patient preferences have been shown to mitigate dropout rates and enhance treatment satisfaction (De Las Cuevas et al., Reference De Las Cuevas, Peñate and de Rivera2014; Windle et al., Reference Windle, Tee, Sabitova, Jovanovic, Priebe and Carr2020). Recognizing this, current guidelines increasingly advocate for the inclusion of patient values and preferences into personalized care. However, patients often encounter barriers, such as internalized stigma, diminished self-efficacy, and cognitive impairment, that impede their ability and willingness to actively engage in decision-making processes or clearly articulate treatment priorities. Consequently, patient preferences are often poorly understood and remain underexplored.
Despite growing interest in patient preferences for MDD treatment, existing literature remains heavily skewed toward treatment-resistant populations or macro-level treatment modalities, such as pharmacotherapy versus psychotherapy (Wittink et al., Reference Wittink, Cary, Tenhave, Baron and Gallo2010; Lokkerbol et al., Reference Lokkerbol, Geomini, van Voorthuijsen, van Straten, Tiemens, Smit, Risseeuw and Hiligsmann2019; Fifer et al., Reference Fifer, Puig, Sequeira, Acar, Ng, Blanchard, Cabrera, Freemantle and Grunfeld2021; dosReis et al., Reference dosReis, Bozzi, Butler, Xie, Chapman, Bright, Malik and Slejko2023). Some inquiries have primarily addressed service-related attributes, such as waiting time and treatment intensity, or risk–benefit trade-offs in highly specific clinical sub-groups (Lokkerbol et al., Reference Lokkerbol, Geomini, van Voorthuijsen, van Straten, Tiemens, Smit, Risseeuw and Hiligsmann2019; Fifer et al., Reference Fifer, Puig, Sequeira, Acar, Ng, Blanchard, Cabrera, Freemantle and Grunfeld2021). However, few studies have explicitly investigated patient preferences for granular pharmacological characteristics, including efficacy, side effects, and cost. Furthermore, current evidence is largely derived from Western cohorts, limiting its generalizability to Chinese patients. Methodologically, while discrete choice experiments (DCE) are well-established, the application of Best-Worst Scaling (BWS) to capture patient preferences in this clinical context remains unexplored (Wouters et al., Reference Wouters, Van Dijk, Van Geffen, Gardarsdottir, Stiggelbout and Bouvy2014; Lokkerbol et al., Reference Lokkerbol, Geomini, van Voorthuijsen, van Straten, Tiemens, Smit, Risseeuw and Hiligsmann2019; Fairchild et al., Reference Fairchild, Katz, Reed, Johnson, Di Bernardo, Hough, Sing and Levitan2020; Fifer et al., Reference Fifer, Puig, Sequeira, Acar, Ng, Blanchard, Cabrera, Freemantle and Grunfeld2021).
As another validated multi-attribute preference assessment method with established formal statistical and measurement properties, BWS has gained increasing traction in healthcare (Flynn et al., Reference Flynn, Louviere, Peters and Coast2007; Flynn, Reference Flynn2010). By presenting respondents with various subsets of attributes, BWS tasks require individuals to select the most and least preferred items. This “best-worst” choice format facilitates a comprehensive ranking of attributes by statistically modeling the relative distance between preferences. Compared to DCE, BWS focuses on the extremes of the preference spectrum, which may theoretically reduce cognitive burden and improve efficiency. It serves as a valuable tool for understanding patient preferences, providing richer information than conventional “pick-one” tasks such as DCE (Flynn et al., Reference Flynn, Louviere, Peters and Coast2007).
Given the high disease burden of MDD and clinical challenges of suboptimal adherence, there is a critical need to better characterize patient preferences for pharmacological treatment. This study employs profile case BWS (BWS-2) to elicit preferences for antidepressant attributes, specifically balancing perceived benefits, risks, and costs, within a Chinese cohort. By quantifying these priorities, the findings are expected to inform shared decision-making between patients and physicians, foster treatment persistence, and improve long-term clinical outcomes for MDD patients.
Methods
Study sample
This study utilized a BWS-2 survey to elicit patient preferences for key antidepressant attributes. Data were collected via convenience sampling across seven psychiatric institutions in Shanghai, China, comprising one tertiary specialized hospital and six secondary-level hospitals. The participating facilities were distributed to include both urban (n = 6) and suburban (n = 2) catchment areas (one with two separate campuses).
Eligible participants included adults with a clinician-confirmed diagnosis of depression, encompassing both first-episode and recurrent cases. Patients were excluded if they had comorbid severe mental disorders, cognitive impairments, or significant physical disabilities; were pregnant; were concurrently participating in other trials; or had a severe depressive episode or unstable mood state, as determined by the attending clinician.
Currently, there is no universally accepted method for determining sample size in BWS studies. Pearmain et al. (Reference Pearmain, Kroes and Bradley1991) have empirically recommended a minimum of 100 participants to support meaningful preference modeling. A systematic review of BWS studies in healthcare reported a median sample size of approximately 162 for BWS-2 experiments (Cheung et al., Reference Cheung, Wijnen, Hollin, Janssen, Bridges, Evers and Hiligsmann2016). Thus, this study targeted a minimum sample size of 300 participants to ensure robust data collection and sufficient statistical power for BWS-2 analysis.
Attributes selection and refinement
The selection of attributes and levels followed a systematic, multi-stage process to ensure clinical relevance, salience and methodological rigor. Initially, a targeted literature review, synthesizing evidence from prior research on efficacy, safety, acceptability and patient preferences associated with antidepressants, supplemented by drug package inserts, identified an 18-attribute candidate pool. These attributes were refined through focus group discussions with a panel of clinical experts (three psychiatrists and two pharmacologists). During this stage, attributes deemed peripheral, indistinct or of low incidence (e.g., mode of administration, dosing frequency, tremor or skin diseases) were excluded, while clinically related attributes were consolidated (e.g., nausea and vomiting) to minimize respondent cognitive burden and avoid decision-making fatigue.
It is often impractical to incorporate too many attributes in BWS due to sample size constraints and the potential reduction in statistical power (De Brún et al., Reference De Brún, Flynn, Ternent, Price, Rodgers, Ford, Rudd, Lancsar, Simpson, Teah and Thomson2018). BWS-1 instrument was developed to derive the relative importance of the candidate attributes among MDD patients and depression-prone groups. It has been demonstrated that BWS-1 can serve as a reliable and quantitatively robust tool to inform attribute selection in stated preference research (Webb et al., Reference Webb, Meads, Lynch, Judge, Randall, Goldbart, Meredith, Moulam, Hess and Murray2021). Drawing on these quantitative results alongside expert interviews, we finalized six key attributes for the BWS-2 instrument: risk of liver/kidney injury, risk of headache/dizziness, sleep disorders, remission rate, risk of gastrointestinal adverse reactions, and monthly out-of-pocket costs. Final refinement of these attributes was conducted in consultation with clinical psychiatrists to ensure clinical relevance and practical applicability.
BWS-2 involves a structured framework of attributes and their corresponding levels, which were established based on a comprehensive literature review and expert consultation. Clinical efficacy and side-effect levels were derived from evidence-based sources, including randomized controlled trials, meta-analyses, reviews, and drug package inserts. Cost levels were informed by public bid prices and negotiated prices under healthcare insurance schemes. To ensure alignment with real-world conditions, patient-reported expenditure data were considered from our pilot survey. Cost levels were originally defined in Chinese yuan (CNY) and were additionally converted into international dollars using purchasing power parity conversion factors obtained from the World Bank (2024). Explanations and visual aids were provided to ensure participant comprehension. The final set of attributes and levels is presented in Table 1.
Explanation of BWS-2 study attributes
Table 1. Long description
The table consists of three columns: Attributes, Explanation, and Levels.
* Remission rate: Defined as the probability of achieving near-normal social functioning or substantial remission of depressive symptoms within 12 weeks. Levels are 25 percent, 55 percent, and 80 percent.
* The risk of headache/dizziness: Defined as the risk induced by antidepressant medication. Levels are 10 percent, 25 percent, and 40 percent.
* Sleep disorders: Defined as the impact of antidepressant medication on sleep. Levels are Somnolence and Insomnia.
* The risk of gastrointestinal adverse reactions: Defined as risks such as nausea, vomiting, diarrhea, and abdominal distension. Levels are 10 percent, 35 percent, and 60 percent.
* The risk of liver/kidney injury: Defined as impaired hepatic and renal function. Levels are 1 percent, 5 percent, and 10 percent.
* Monthly out-of-pocket costs: Defined as expenses after medical insurance reimbursement. Levels are 100 C N Y (U S dollar 28), 350 C N Y (U S dollar 99), and 600 C N Y (U S dollar 170).
A footnote specifies that C N Y stands for Chinese Yuan.
CNY: Chinese Yuan.
Experimental design
The BWS-2 choice sets were generated using an orthogonal main effects design (OMED) via SAS 9.4. This design ensures both balance (equal frequency of attribute levels) and orthogonality (equal frequency of attribute level pairs), thereby optimizing design efficiency (Reed Johnson et al., Reference Reed Johnson, Lancsar, Marshall, Kilambi, Muhlbacher, Regier, Bresnahan, Kanninen and Bridges2013). The structure enables the independent estimation of preference weights for each attribute level, maximizing the robustness of the analysis (Louviere et al., Reference Louviere, Flynn and Marley2015).
The experimental design included 18 choice sets, each featuring six attributes and their respective levels. To minimize cognitive burden and survey fatigue, these sets were partitioned into three blocks of six tasks each. To verify internal validity and ensure participant comprehension of the BWS-2 format, a quality control task was embedded. Participants were presented with a single attribute (the risk of headache/dizziness) across six levels (0 to 60%) and expected to select 0 as the best and 60% as the worst option. Participants were randomly assigned to one block and completed six tasks along with this validity test. A representative BWS-2 task is illustrated in Figure 1.
An example of a BWS-2 task.
Figure 1. Long description
A survey table titled ‘Considering the features below, which is the best and the worst when choosing/taking antidepressants?’. The table consists of three columns. The left column is headed ‘Which is the best feature of this antidepressant?’. The center column is headed ‘Characteristics of antidepressants’. The right column is headed ‘Which is the worst feature of this antidepressant?’.
There are six rows of data, each containing a central characteristic and a radio button in both the left and right columns for user selection. The characteristics are as follows:
1. The risk of liver/kidney injury: 5%
2. The risk of headache/dizziness: 35%
3. The risk of gastrointestinal adverse events: 60%
4. Sleep disorders: Insomnia
5. Remission rate: 80%
6. Monthly out-of-pocket costs: 600 C N Y
Data collection
The survey was administered face-to-face across seven psychiatric institutions in Shanghai, China. To ensure data quality, investigators underwent training on the study objectives, questionnaire content, and the BWS-2 task structure. Prior to the choice tasks, participants were provided with standardized attribute definitions and practice examples.
Eligibility for survey participation was determined by attending clinicians, who assessed patients for clinical stability, cognitive capacity, and ability to make informed choices. They completed the questionnaire anonymously after providing written informed consent. To ensure data integrity, investigators monitored the process, with clinical staff present to build rapport and facilitate patient cooperation. In instances where participants encountered difficulties, such as geriatric patients or those with lower education levels, investigators conducted oral inquiries to ensure response validity and completeness.
Real-time communication with the project coordination team was facilitated to promptly address any issues. Upon completion, investigators reviewed the questionnaires with participants to rectify potential omissions. Surveys were terminated immediately if a participant exhibited signs of distress or cognitive fatigue, as evaluated by the healthcare team. All participants received compensation for their time and effort.
Statistical analysis
Data were double-entered into EpiData 3.1 to ensure accuracy, and subsequent analysis was performed using STATA 17.0 and RStudio. Descriptive statistics were used to summarize the sociodemographic and health-related characteristics.
We employed the marginal sequential method to analyze BWS responses, which assumes that respondents first select the best option from J alternatives, followed by the worst from the remaining J-1 alternatives; this approach aligns with random utility theory. To estimate preference weights, we applied conditional logit (CLM), mixed logit (MIXL) and generalized multinomial logit (GMNL) models (Fiebig et al., Reference Fiebig, Keane, Louviere and Wasi2010; Muhlbacher et al., Reference Muhlbacher, Zweifel, Kaczynski and Johnson2016). While the CLM assumes preference homogeneity, the MIXL and GMNL models allow for preference heterogeneity by assuming a distribution, typically normal, defined by mean and standard deviation (Train and Sonnier, Reference Train and Sonnier2003; Fiebig et al., Reference Fiebig, Keane, Louviere and Wasi2010; Clark et al., Reference Clark, Determann, Petrou, Moro and de Bekker-Grob2014). Model fit was evaluated via the Akaike (AIC) and the Bayesian (BIC) information criterion. All attributes were dummy-coded, with “Insomnia” designated as the reference level. Model coefficients represent relative utilities. Positive values signify higher utility (greater preference), while negative values denote lower utility relative to the reference level. Statistical significance was set at p < 0.05.
Attribute importance weights were calculated by measuring the range of part-worth utilities, defined as the difference between the minimum and maximum coefficients within each attribute (Weernink et al., Reference Weernink, van Til, Groothuis-Oudshoorn and IJ2017; Jiang et al., Reference Jiang, Gu, Yang, Wu, Wang, Cutler and Zhang2020). These weights were normalized to a 0–1 scale by dividing the utility range of each attribute by the largest observed value, assigning a weight of 1 to the most impactful attribute.
Finally, exploratory subgroup analyses (stratified by age, gender, education, and income) were conducted to identify potential sources of preference heterogeneity and understand preference variations across diverse populations.
Results
Participant characteristics
A total of 356 patients were enrolled across seven mental health centers. After excluding 25 participants due to incomplete BWS tasks (n = 7), failure to meet inclusion criteria (n = 10), or failing the quality control test (n = 8), 331 valid responses were retained for analysis.
The study cohort had a mean age of 54.85 ± 19.00 years. The majority were female (67.67%), married (65.26%), and urban-dwelling (93.35%). Around 67.04% completed senior or higher education. Over half were retired (50.76%) and reported an annual income or pension below 60,000 CNY (US$16,997, 61.03%). Consistent with national healthcare trends, the vast majority (> 95%) were enrolled in medical insurance schemes (National Healthcare Security Administration, 2024).
Clinically, more patients were in the recurrent treatment stage (58.01%) and received pharmacological therapy only (60.73%). A majority had been on antidepressant medication for over two years (62.24%), with monthly medication costs not exceeding 400 CNY (US$113, 75.22%). Further details on sociodemographic and clinical characteristics are presented in Table 2.
Sociodemographic and health-related characteristics of the respondents
Table 2. Long description
The table presents data for a total sample of N = 331.
Left Column Characteristics:
* Gender: Male 106 (32.02%), Female 224 (67.67%), Missing 1 (0.03%).
* Age: Mean 54.85 (S D 19.00, range 18 to 90). Groups include 18 to 44 years (32.63%), 45 to 59 years (16.31%), 60 to 74 years (35.95%), and 75 years or older (14.50%).
* Residence: Rural (6.65%), Urban (93.35%).
* Marital status: Single (21.45%), Married (65.26%), Divorced or separated (6.34%), Widowed (6.95%).
* Education: Junior or below (32.63%), Senior (29.61%), College and above (37.46%).
* Employment status: Retired is the largest group at 50.76%, followed by Stable jobs (14.50%) and Students (10.88%).
* Annual personal income in C N Y: The largest group earns 20,000 to 60,000 (43.81%).
Right Column Characteristics:
* Types of depression: Recurrent depression (42.30%), Anxiety disorder with depression (22.05%), Mild (9.67%), Moderate (9.67%), Bipolar disorder (7.55%), and Severe (7.25%).
* Treatment stage: Initiate treatment (38.67%), Recurrent treatment (58.01%), No treatment (3.32%).
* Treatment mode: Drug therapy (60.73%), Combination therapy (35.35%), Psychotherapy (0.60%).
* Health insurance: Yes (95.76%), No (2.42%).
* Duration of medication: Greater than 24 months (62.24%), 1 to 6 months (14.20%), 12 to 24 months (9.97%).
* Monthly out-of-pocket costs in C N Y: Distributed relatively evenly across brackets from less than 50 to greater than 400, with each bracket representing approximately 17% to 20% of the sample.
CNY: Chinese Yuan; SD: Standard deviation.
Results of BWS analysis
The improvement in model fit was reflected by enhanced log-likelihood and improved AIC and BIC values, indicating that the MIXL outperformed the GMNL, with the CLM being the least effective. Consistent estimates were observed across all models, detailed in Supplementary Appendix 1. Thus, results from the MIXL are reported herein.
The MIXL model estimates, presented in Table 3, were anchored to the reference level “insomnia,” which was consistently rated as a less preferred attribute level. With the exception of the 60% risk of gastrointestinal adverse reactions, all other attribute levels yielded positive mean preference weights, indicating higher desirability compared to “insomnia.” All attribute levels, except for the 35% (β = 0.07, P = 0.512) and 60% (β = −0.19, P = 0.091) risks of gastrointestinal adverse reactions, demonstrated statistically significant preference weights (P<0.05). The significant standard deviations in the MIXL model also indicated potential preference heterogeneity, confirming variability in treatment priorities across the study population.
Results of mixed logit model
Table 3. Long description
The table contains nine columns: Attributes and levels, Coefficient, S E, P, S D, S E, P, Rank, and Attribute impact.
Key attributes and their data include:
* Risk of liver/kidney injury: Levels 1%, 5%, and 10% have coefficients of 1.65, 1.39, and 1.12 respectively, with an attribute impact of 0.098.
* Risk of headache/dizziness: Levels 10%, 25%, and 40% have coefficients of 0.84, 0.42, and 0.24 respectively, with an attribute impact of 0.112.
* Risk of gastrointestinal adverse reactions: Levels 10%, 35%, and 60% have coefficients of 0.59, 0.07, and minus 0.19 respectively, with an attribute impact of 0.144.
* Sleep disorders: Somnolence has a coefficient of 0.77, while Insomnia is the reference level. Attribute impact is 0.143.
* Remission rate: Levels 80%, 55%, and 25% have coefficients of 6.06, 3.12, and 0.66 respectively. This attribute has the highest impact at 1.000.
* Monthly out-of-pocket costs: Levels 100 C N Y, 350 C N Y, and 600 C N Y have coefficients of 1.64, 1.36, and 1.01 respectively, with an attribute impact of 0.116.
Model fit statistics at the bottom:
* Log likelihood: minus 5,224.
* A I C: 10,513.
* B I C: 10,769.
* Respondents n: 331.
* Observations n: 21,846.
SD: Standard deviation; SE: Standard error; AIC: Akaike information criterion; BIC: Bayesian information criterion; CNY: Chinese Yuan.
BWS-2 enabled a holistic ranking of all attribute levels based on their relative utilities. Among patients, the 80% (β = 6.06, P<0.05) and 55% (β = 3.12, P<0.05) remission rates were associated with the highest utility, followed by the 1% risk of liver/kidney injury (β = 1.65, P<0.05) and a monthly out-of-pocket cost of 100 CNY (US$28, β = 1.64, P<0.05). These results indicated that patients prioritized antidepressants with higher efficacy, lower risk of liver/kidney injury, and lower costs when selecting antidepressants. Conversely, the 60% risk of gastrointestinal adverse reactions (β = −0.19, P = 0.091), insomnia, and the 35% risk of gastrointestinal adverse reactions (β = 0.07, P = 0.512) were ranked as less desirable, followed by the 40% (β = 0.24, P = 0.04) and 25% (β = 0.42, P<0.05) risks of headache/dizziness.
The relative importance weights for each attribute, derived from the MIXL utility range, are presented in Table 3. Among all attributes, “remission rate” exerted the most substantial impact on treatment choice, with a utility range of 5.4 (relative to an 80% remission rate). The ranking of attributes in descending order of relative importance was as follows: remission rate, risk of gastrointestinal adverse reactions, sleep disorders, monthly out-of-pocket costs, risk of headache/dizziness and risk of liver/kidney injury. These results underscore that the remission rate is the most significant factor in antidepressant selection, followed by concerns about side effects and financial costs associated with treatment.
Subgroup analysis
Figure 2 illustrates the comparative results among subgroups based on the MIXL model. Despite variations in attribute prioritization, remission rate consistently emerged as the most important driver of patient selection. Subgroup analysis revealed nuanced preference patterns. Female prioritized sleep disorders, risks of gastrointestinal adverse reactions and headaches/dizziness, while men were more influenced by monthly out-of-pocket costs and risk of liver/kidney injury. Elderly patients (≥60 years) were more concerned with sleep disorders, monthly out-of-pocket costs, and risk of headaches/dizziness. Conversely, younger patients prioritized risks of liver/kidney injury and gastrointestinal adverse reactions. Higher educational attainment correlated with a lower emphasis on monthly out-of-pocket costs and sleep disorders, alongside a higher priority for risk of gastrointestinal adverse reactions. Patients receiving initial treatment were more concerned with risk of liver/kidney injury, while those undergoing retreatment placed greater weight on sleep disorders and risk of gastrointestinal adverse reactions. Patients on combined therapy prioritized risk of liver/kidney injury, while those on monotherapy emphasized sleep disorders and risk of headaches/dizziness. Patients with lower monthly out-of-pocket costs emphasized these costs and sleep disorders, whereas those facing higher costs prioritized risks of headaches/dizziness and gastrointestinal adverse reactions. Detailed comparative results are shown in Supplementary Appendix 1.
The relative importance of attributes among subgroups.
Figure 2. Long description
The chart consists of horizontal bars grouped into seven categories. Each bar is composed of six colored segments representing different attributes. From left to right, the attributes are: Risk of liver/kidney injury (dark blue), Risk of headache/dizziness (medium blue), Risk of gastrointestinal adverse reactions (teal), Sleep disorders (light green), Remission rate (pale green), and Monthly out-of-pocket (light orange).
Key data points by category:
* Gender: Males prioritize liver/kidney injury at 12.00% compared to 6.00% for females. Remission rate is the largest segment for both at 58.00% and 60.00% respectively.
* Age: Those over 60y show a higher concern for sleep disorders (14.00%) than those below 60y (6.00%).
* Education: The College or above group has the highest remission rate importance at 71.00% and the lowest monthly out-of-pocket importance at 3.00%.
* Annual income: The Over 60000 group places 76.00% importance on remission rate, significantly higher than the Below 60000 group at 52.00%.
* Treatment stage: Recurrent patients value remission rate (64.00%) more than initial patients (55.00%).
* Treatment mode: Only pharmacotherapy shows a slightly higher remission rate importance (62.00%) than combined treatment (59.00%).
* Monthly O O P (C N Y): The Below 100 group has the highest sleep disorder concern at 16.00%, while the Over 400 group has the highest monthly out-of-pocket concern at 9.00%.
Discussion
To our knowledge, this study represents the first application of the BWS-2 instrument to investigate patient preferences for antidepressants. Given the profound disease burden and the persistent challenge of suboptimal adherence, integrating patient-centered perspectives into the clinical decision-making process is essential. Our findings delineate patient priorities in antidepressant selection, aimed at providing robust, evidence-based insights to inform clinical practice, foster treatment persistence and improve long-term outcomes.
High efficacy, defined by remission rate, serves as the most desirable characteristic of antidepressants and the paramount driver of patient choice. This finding aligns with previous DCE research on treatment-resistant populations (Fairchild et al., Reference Fairchild, Katz, Reed, Johnson, Di Bernardo, Hough, Sing and Levitan2020; Fifer et al., Reference Fifer, Puig, Sequeira, Acar, Ng, Blanchard, Cabrera, Freemantle and Grunfeld2021). Our results also corroborate Wouters et al. (Reference Wouters, Van Dijk, Van Geffen, Gardarsdottir, Stiggelbout and Bouvy2014), who observed that relapse prevention and symptom relief were equally valued by patients. This suggests a willingness among patients to potentially tolerate certain side effects in exchange for a higher probability of remission. Since evidence suggests that the efficacy of commonly used first-line antidepressants is often perceived as comparable (Cipriani et al., Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa, Leucht, Ruhe, Turner, Higgins, Egger, Takeshima, Hayasaka, Imai, Shinohara, Tajika, Ioannidis and Geddes2018), considering patient-preferred attributes becomes essential in treatment selection. Notably, the relative influence of efficacy on patient choice can be sensitive to its operationalization. For instance, defining efficacy through clinical remission rates versus other metrics or varying the range of attribute levels beyond the bounds of patients’ expectations may alter patient priority. This methodological sensitivity highlights the need for careful attribute framing in preference research.
High risk of gastrointestinal (GI) adverse reactions was identified as the least desirable feature of antidepressants. This observation is consistent with prior conjoint analysis research (Wouters et al., Reference Wouters, Van Dijk, Van Geffen, Gardarsdottir, Stiggelbout and Bouvy2014). GI symptoms, such as nausea, diarrhea, and constipation, represent some of the most common antidepressant-related side effects, frequently precipitating patient discomfort and potential treatment discontinuation(Ho et al., Reference Ho, Jacob and Tangiisuran2017; Oliva et al., Reference Oliva, Lippi, Paci, Del Fabro, Delvecchio, Brambilla, De Ronchi, Fanelli and Serretti2021; Jia et al., Reference Jia, Yiyun, Zhiguo, Yousong, Min, Yifan, Na, Feng, Yiru and Daihui2022). These reactions may reflect underlying physiological disruptions, such as gut microbiome alterations and potential bioaccumulation, which can impair bioavailability and attenuate therapeutic efficacy (Chen et al., Reference Chen, Park, Li and LE2020; Manchia et al., Reference Manchia, Fontana, Panebianco, Paribello, Arzedi, Cossu, Garzilli, Montis, Mura, Pisanu, Congiu, Copetti, Pinna, Pazienza, Squassina and Carpiniello2021). We categorized these symptoms into a unified attribute to minimize respondent cognitive burden, as BWS designs become statistically inefficient when incorporating excessive attributes. However, we acknowledge that this compromise may mask symptom-specific utility weights. Given that clinical impacts often diverge (Oliva et al., Reference Oliva, Lippi, Paci, Del Fabro, Delvecchio, Brambilla, De Ronchi, Fanelli and Serretti2021), future studies can explore these granular attributes to refine personalized clinical decision-making. Nonetheless, as GI tolerability remains a critical concern, it is essential to integrate GI assessment into treatment decisions.
Sleep disorders were categorized into two distinct types: insomnia and somnolence, differing from prior research that often focused on varying severity levels or solely on insomnia (Zimmermann et al., Reference Zimmermann, Clouth, Elosge, Heurich, Schneider, Wilhelm and Wolfrath2013; Ng-Mak et al., Reference Ng-Mak, Poon, Roberts, Kleinman, Revicki and Rajagopalan2018). This dichotomization clarified patient preferences regarding these prevalent adverse effects while minimizing cognitive burden. Both conditions were regarded as undesirable, reflecting their status as the most frequent sleep-related side effects (Natter et al., Reference Natter, Yokoyama and Michel2021; Zhou et al., Reference Zhou, Li, Yang, Xiong and Xiong2023a). Given that pharmacological profiles often favor either insomnia (e.g., reboxetine, vilazodone) or somnolence (e.g., fluvoxamine, trazodone), this assessment is clinically relevant (Wichniak et al., Reference Wichniak, Wierzbicka, Walecka and Jernajczyk2017; Zhou et al., Reference Zhou, Li, Lv, Lai, Liu, Zhou, Liu, Tao, Zhang, Yu, Tian and Sun2023b). This study revealed that sleep-related preferences are influenced by some characteristics, particularly among females, older patients, those on monotherapy, and those in relapse treatment. The findings may offer valuable insights for informing personalized treatments and promoting effective clinician-patient communication.
While low monthly out-of-pocket costs were considered a desirable antidepressant feature, this attribute exerted a moderate influence on patient choices. This finding contrasts with Fifer et al. (Reference Fifer, Puig, Sequeira, Acar, Ng, Blanchard, Cabrera, Freemantle and Grunfeld2021), where cost was a more important driver for Australian treatment-resistant patients, likely due to their broader cost range ($100–$2,500) and different clinical context. Additionally, patients in that research might be concerned about side effects over other attributes, such as dosing frequency or time to response, which underscores the importance of clinically relevant and meaningful attribute selection. Notably, our sample may underrepresent vulnerable populations, specifically those with financial hardship or lacking medical insurance. Since high national health insurance coverage likely buffers cost sensitivity, this potentially limits the generalizability of our findings to populations facing severe financial barriers. Given that financial stress remains a catalyst for non-adherence, clinicians should account for treatment costs and patients’ financial capacities when adjusting medication regimens.
Although the low risk of liver/kidney injury was considered a desirable antidepressant characteristic, its direct impact on patient choices remained limited. This attribute, often overlooked in previous studies due to the historically low incidence of drug-induced liver injury (DILI) or insufficient evidence, was incorporated here following expert consultations and findings from BWS-1(Friedrich et al., Reference Friedrich, Akimova, Huf, Konstantinidis, Papageorgiou, Winkler, Toto, Greil, Grohmann and Kasper2016; Damba et al., Reference Damba, Bodenstein, Lavin, Drury, Sekhon, Renoux, Trinh, Rej and Greenway2022; Khalil et al., Reference Khalil, MacKenzie, Maletic-Savatic and Li2024). Recent evidence indicates a rise in reported antidepressant-associated DILI, particularly cases involving life-threatening events or hospitalizations (Carvalho et al., Reference Carvalho, Sharma, Brunoni, Vieta and Fava2016; Jiang et al., Reference Jiang, Wei, Zhu, Wu and Wu2024). Collectively, these trends underscore the growing need for heightened awareness of DILI risks.
The relatively low importance of liver/kidney injury risk likely stems from several factors. First, the narrow risk range examined (1%–10%) may have limited its perceived significance relative to other attributes. Second, patients often exhibit a cognitive bias toward immediate, symptomatic discomfort over low-probability, latent safety risks. Essentially, patients prioritize the mitigation of tangible, present-day burdens over the avoidance of distant medical threats. Nonetheless, younger, treatment-naive, and combination-therapy patients expressed greater concern, likely reflecting heightened long-term safety awareness. Given the prolonged nature of MDD treatment, it remains a critical consideration for high-risk profiles, such as the elderly or those with chronic comorbidities, where regular monitoring and patient-centered counseling regarding latent risks are essential.
Beyond identifying specific preferences, our findings provide a concrete evidence-based foundation for pre-prescription counseling. These empirical data offer clinicians actionable ‘talking points’ to facilitate shared decision-making and manage patient expectations. By fostering a transparent discussion of these attributes, clinicians can strengthen the therapeutic alliance, thereby empowering patients to overcome the inertia associated with MDD and engage more actively in their treatment journey. Although our study utilized subgroup analyses to explore preference variations, real-world treatment decisions are complex. A potential direction is to further investigate how diverse patient profiles might moderate the relative importance of clinical and financial attributes. Such investigations, which could benefit from larger, more socio-demographically diverse cohorts and potentially broader experimental designs, could contribute to further elucidating these complex trade-offs. Moreover, it is essential to recognize that patient preferences for antidepressant attributes may be influenced by their sociocultural environment. While our BWS model focuses on drug-specific features, sociocultural factors like stigma may act as a ‘hidden’ moderator which is difficult to quantify. Future studies can explore how social stigma moderates the trade-offs patients make during treatment selection.
Limitations
This study has several limitations. First, sampling was affected by COVID-19-related mobility restrictions, potentially limiting the representativeness of our cohort, as the sample mainly comprised urban residents with relatively high educational attainment and mean age in economically developed regions. While we conducted subgroup analyses to mitigate these disparities, the underrepresentation of vulnerable groups remains a constraint. Future research should encompass samples from diverse geographic regions and social strata. Second, to restrict the number of attributes and prevent cognitive overload, certain side effects were consolidated. While necessary for survey efficiency, this trade-off may mask specific preference weights for individual symptoms.
Conclusion
This study provides the first empirical evidence on antidepressant preferences among Chinese patients using the BWS-2 instrument. Our findings identify remission rate as the most important consideration, followed by avoidance of specific side effects, particularly gastrointestinal symptoms and sleep disorders, and treatment costs. Variations in preferences were observed across sociodemographic and clinical subgroups, underscoring the need for further exploration. By incorporating patient-centered priorities into the clinical dialog, providers can optimize personalized treatment plans, improve adherence, and ultimately enhance long-term clinical outcomes.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/gmh.2026.10287.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/gmh.2026.10287.
Acknowledgments
The authors would like to thank all survey participants.
Author contribution
S.L. and Y.C. conceptualized and obtained funding for this research. S.L. and Y.C. contributed to the design of the study. F.X., H.S., S.B., S.L. and Y.C. advised on the research as people with expertise by experience. Y.R., Y.T., F.X., H.S., S.B., S.L. and Y.C. all took part in some aspects of the data collection. Y.T. and Y.R. prepared and analyzed the data. Y.T. wrote the first draft of the paper, with supervision from S.L. and Y.C., with assistance from Y.R., F.X., H.S., S.B., S.L. and Y.C. revised the manuscript and provided critically important intellectual content. All authors had an opportunity to comment on the final article. Y.C. and S.L. are the guarantors responsible for the overall content. All authors read and approved the final manuscript.
Declaration of interest
None declared.
Data availability statement
Data are available on request. The data sets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Ethics and approvals
This study was approved by the Medical Research Ethics Committee of the School of Public Health, Fudan University, which approved the study on May 3, 2022 (Reference No. IRB#2021-2107-0911-S1), and the research adhered to the principles of the Declaration of Helsinki. Following a complete study description, all patients gave informed consent prior to their inclusion.
Funding statement
This work was funded by the National Natural Science Foundation of China Young Scientists Fund (Grant No. 72404059) and National Natural Science Foundation of China General Program (Grant No. 72074047). The funding source had no involvement in the conduct of the research and/or preparation of the article.
Patient and public involvement
Patients and/or the public were involved in the conduct, reporting or dissemination plans of this research. Refer to the Methods section for further details.
Patient consent for publication
Not applicable.
Open access
