Introduction
Both bipolar disorder (BD) and major depressive disorder (MDD) are highly prevalent mental disorders associated with substantial disability and functional impairment, increased suicidality, and high mortality (GBD 2015 Disease and Injury Incidence and Prevalence Collaborators, 2016; McIntyre et al., Reference McIntyre, Berk, Brietzke, Goldstein, Lopez-Jaramillo, Kessing, Malhi, Nierenberg, Rosenblat, Majeed, Vieta, Vinberg, Young and Mansur2020). It was found that a large proportion (58.9%) of first episodes of BD were depressive (Baldessarini, Tondo, & Visioli, Reference Baldessarini, Tondo and Visioli2014), and in the early stage of BD, patients often present with symptoms resembling those of MDD (Zhu et al., Reference Zhu, Kou, Mu, Hu, Yuan, Yuan, Luo and Zhang2025). Currently, internationally recognized diagnostic criteria for mental disorders, such as those in the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Diseases, are still based on observable symptoms. As a result, many patients with early stage BD are diagnosed with and treated for MDD.
Previous studies have found that patients with BD may benefit from interventions as early as possible (Kessing et al., Reference Kessing, Hansen, Hvenegaard, Christensen, Dam, Gluud and Wetterslev2013; Ratheesh et al., Reference Ratheesh, Hett, Ramain, Wong, Berk, Conus, Fristad, Goldstein, Hillegers, Jauhar, Kessing, Miklowitz, Murray, Scott, Tohen, Yatham, Young, Berk and Marwaha2023). However, for patients with early stage BD who received a diagnosis of MDD, appropriate treatments may be delayed. Also, as conventional antidepressant monotherapy for MDD is found to be associated with various detrimental outcomes for patients with BD (Pacchiarotti et al., Reference Pacchiarotti, Bond, Baldessarini, Nolen, Grunze, Licht, Post, Berk, Goodwin, Sachs, Tondo, Findling, Youngstrom, Tohen, Undurraga, González-Pinto, Goldberg, Yildiz, Altshuler and Vieta2013; Viktorin et al., Reference Viktorin, Lichtenstein, Thase, Larsson, Lundholm, Magnusson and Landen2014), a delayed detection of the diagnostic conversion from MDD to BD may not only result in delayed treatment of BD but also prolong the patients’ exposure to unsuitable treatment, possibly leading to worse outcomes for the patients. Therefore, detecting the diagnostic conversion from MDD to BD as early as possible is crucial for timely treatment and a better prognosis. Previous studies found that the rate of diagnostic conversion to BD varied from 2.82% to 19.0% (Kung et al., Reference Kung, Tsai, Cheng, Tsai, Su, Chen, Bai, Liang and Chen2024; Service et al., Reference Service, De La Hoz, Diaz-Zuluaga, Arias, Pimplaskar, Luu, Mena, Valencia-Echeverry, Ramírez, Bearden, Sabatti, Reus, López-Jaramillo, Freimer and Olde Loohuis2025; Zhu et al., Reference Zhu, Kou, Mu, Hu, Yuan, Yuan, Luo and Zhang2025). The high conversion rate means that a large number of patients may benefit from the early detection of diagnostic conversion to BD. Hence, identifying predictors associated with the conversion to BD is of great clinical significance.
Earlier studies have identified various predictors for the conversion to BD, including demographic characteristics (such as sex or age), clinical characteristics (such as comorbidities or severity or recurrence of depression), and differentiating markers (such as patterns of neuronal activities (Kung et al., Reference Kung, Tsai, Cheng, Tsai, Su, Chen, Bai, Liang and Chen2024; Service et al., Reference Service, De La Hoz, Diaz-Zuluaga, Arias, Pimplaskar, Luu, Mena, Valencia-Echeverry, Ramírez, Bearden, Sabatti, Reus, López-Jaramillo, Freimer and Olde Loohuis2025; Sun et al., Reference Sun, Yan, Hua, Xia, Chen, Huang, Wang, Xia, Yao and Lu2024; Zhu et al., Reference Zhu, Kou, Mu, Hu, Yuan, Yuan, Luo and Zhang2025). However, it is difficult to translate those factors into clinical practice. Also, those previously identified factors were static characteristics, and to date, the dynamic in the patients’ illness course, specifically the stability and fluctuation of patients’ conditions, has not been thoroughly investigated as predictors of the conversion to BD. BD is characterized by fluctuations in mood state (Grande, Berk, Birmaher, & Vieta, Reference Grande, Berk, Birmaher and Vieta2016). Previous research on the illness course of BD has found that the severity of both depressive and manic/hypomanic symptoms of the patient fluctuated over time (Judd et al., Reference Judd, Akiskal, Schettler, Endicott, Maser, Solomon, Leon, Rice and Keller2002, Reference Judd, Akiskal, Schettler, Coryell, Endicott, Maser, Solomon, Leon and Keller2003). The fluctuating nature of this disease suggests that the dynamic status of symptoms may be an under-explored yet crucial aspect in the research on the conversion to BD.
Time in target range (TTR) is defined as the proportion of time a given index stays within a predefined target range. To date, the concept of TTR has been used in studies on various chronic diseases that require long-term management to quantify the stability of relevant conditions (Jacobs et al., Reference Jacobs, Woller, Stevens, May, Bair, Anderson, Crandall, Day, Johanning, Long, Mallender, Olson, Osborn, Weiss and Bunch2014; Wang et al., Reference Wang, Song, Liu, Yu, Wang, Qin and Wu2025; Zhang et al., Reference Zhang, Xu, Liu, Lyu, Bian, Song, Liu, Ma, Liu, Chen and Li2025). Statistical Process Control (SPC) is a useful concept for monitoring and identifying shifts in ever-changing processes (Benneyan, Lloyd, & Plsek, Reference Benneyan, Lloyd and Plsek2003). Previous studies have applied SPC to detect structural mean shifts in various factors and determine the most sensitive early sign of the recurrence of depression (Schreuder et al., Reference Schreuder, Schat, Smit, Snippe and Ceulemans2024; Snippe et al., Reference Snippe, Smit, Kuppens, Burger and Ceulemans2023). In the hope of filling the aforementioned gap in the research on predictors of the conversion to BD and offering reference to clinical practice, the current study aims to investigate the relationships between the dynamic in illness course (specifically, the stability and fluctuation of depressive and manic symptoms), evaluated based on the concepts of TTR and SPC, of patients initially diagnosed with MDD and the diagnostic conversion to BD in those patients (Figure 1).
Graphic abstract.
Figure 1. Long description
The graphic is divided into two main columns.
Left Column:
1. Data Source: The Prospective Cohort Study of Depression (P ROUD).
2. Symptom Assessment: Field visits using HAMD-17, QIDS-SR16, and HCL-33. Remote visits using MDQ and PHQ-9.
3. Outcome: Conversion to bipolar disorder (B D) involving 2367 patients with MD D, with 69 converting in 52 weeks.
4. Index 1 (T TR): Time under the target threshold. Two line graphs show total scores over 52 weeks. A yellow window represents the calculation period (weeks 0-8 for field, 14-22 for remote) and a blue window represents the incidence of conversion (weeks 8-52 or 22-52).
5. Index 2 (S PC): Time in the range of P25th-P75th. Similar graphs show scores within a shaded blue horizontal band representing the interquartile range.
Right Column:
1. Table 1: Negative associations between TTR and conversion to BD. HCL-33 shows significant P values of 0.001 and less than 0.001 for cutoff values of 7 and 15.
2. Table 2: Negative associations between SPC and conversion to BD. HCL-33 shows a significant HR of 0.019 and P value less than 0.001.
3. Subgroup Analysis: A forest plot for SPC and HCL-33 shows Hazard Ratios across categories like first episode of MD D, onset age, family history, and sex.
Bottom Section:
Conclusions: Continuous monitoring of hypomanic symptoms, especially via HCL-33, is clinically significant for early identification of conversion to B D.
Methods
Study design and participants
This study used the data from the Prospective Cohort Study of Depression in China (PROUD) to investigate the relationships between the fluctuation of depressive and manic symptoms and the diagnostic conversion to BD in patients initially diagnosed with MDD. The data analyzed in this study were collected from 20 tertiary psychiatric hospitals from May 2022 to February 2025. More details of the PROUD study were described in its published protocol (Zhou et al., Reference Zhou, Xu, Liu, Qi, Yang, Guo, Zhou, Zhu, Zhang, Chen, Lyu, Feng, Zhang, Liu, Wang, Wang, Zhang, Xiao, Feng and Wang2023). The PROUD study was registered in the Chinese Clinical Trial Registry platform (registration number: ChiCTR2200059053), and the ethical approval for it was obtained from Beijing Anding Hospital, Capital Medical University, Beijing, China (Approval Number 14. [2022] keyan-20221FS-2). All participants have given written informed consents.
In this study, patients were included if they (1) met the diagnosis criteria of MDD in the DSM, 5th Edition (DSM-5); (2) were aged 18–65 years old (no matter men or women); (3) had a total score of 14 or more on the 17-item Hamilton Depression Rating Scale (HAMD-17) at baseline; (4) had not taken any antidepressants for at least 14 days before screening; (5) were planned to receive antidepressants monotherapy; and (6) had completed at least two field visits. Patients were excluded if they had depressive episodes secondary to other diseases, received physical therapy (modified electroconvulsive therapy, transcranial magnetic stimulation, deep brain stimulation, or vagus nerve stimulation) or participated in any interventional clinical trial within 3 months before screening, or (for women) were pregnant or planning to become pregnant at the time of enrollment.
Follow-up regimen
Three types of follow-up visits were included in this study: field visits, remote visits, and unscheduled visits. All patients underwent the same follow-up regimen. The field visits included face-to-face interviews conducted at baseline, Weeks 2, 8, 12, 24, and 52. Remote visits included a remote follow-up checklist (RFC) delivered to the patients through mobile devices every two weeks between Weeks 12 and 52 (this study included data from 17 remote visits – Weeks 14, 16, 18, 20, 22, 26, 28, 30, 32, 34, 38, 40, 42, 44, 46, 48, and 50). The RFC consisted of validated scales and questions about the patient’s condition, amounting to seven items. Items 1 (feeling nervous, anxious, or on edge) and 2 (not being able to stop or control worrying) were from the Generalized Anxiety Disorder-2 (GAD-2), a two-item scale for fast-screening of generalized anxiety disorder symptoms (Kroenke et al., Reference Kroenke, Spitzer, Williams, Monahan and Lowe2007). Items 3 (little interest or pleasure in doing things) and 4 (feeling down, depressed, or hopeless) were from the Patient Health Questionnaire-2 (PHQ-2), a two-item scale assessing depressed mood and anhedonia which can be used to identify patients for further evaluation with the full PHQ-9 (PHQ-9) (Kroenke, Spitzer, & Williams, Reference Kroenke, Spitzer and Williams2003; Levis et al., Reference Levis, Sun, He, Wu, Krishnan, Bhandari, Neupane, Imran, Brehaut, Negeri, Fischer, Benedetti, Thombs, Che, Levis, Riehm, Saadat, Azar and Zhang2020). Items 5 (‘Have thoughts that you would be better off dead, or of hurting yourself in some way’); 6 (‘Has there been a period when you felt very happy, in high spirits, excited or overjoyed?’); and 7 (‘Have there been several days when you felt irritable, angry, or short-tempered for most of the day?’) were questions about the patient’s condition that the researchers identified, based on clinical experience, as signs requiring further evaluation. For further evaluation, the PHQ-9 would be delivered to the patient if the GAD-2 score (the sum of scores of Items 1 and 2) or the PHQ-2 score (the sum of the scores of Items 3 and 4) was 3 or more. The Mood Disorder Questionnaire (MDQ) would be delivered if the answer to Item 6 or 7 was ‘yes’. The unscheduled visits were on-site field visits elicited when the results of those further evaluations met prespecified criteria. In the elicited unscheduled visits, participants would be reevaluated according to the DSM-5 criteria and rediagnosed, and relevant data would be collected. Other evaluations elicited by responses to the RFC (that were irrelevant to the current study) and specific criteria for elicitation of unscheduled visits (named as ‘external visits’ in the protocol) were introduced in the protocol of the PROUD study (Zhou et al., Reference Zhou, Xu, Liu, Qi, Yang, Guo, Zhou, Zhu, Zhang, Chen, Lyu, Feng, Zhang, Liu, Wang, Wang, Zhang, Xiao, Feng and Wang2023).
Outcome measures
The outcome of this study was the manic or hypomanic episode that met the DSM-5 criteria (assessed with the clinician-rated Mini-International Neuropsychiatric Interview, M.I.N.I.-CR) (Sheehan et al., Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs, Weiller, Hergueta, Baker and Dunbar1998) during the field visits and unscheduled visits within 52 weeks.
Clinical assessments
In the field visits and unscheduled visits, the severity of depressive symptoms was evaluated by the HAMD-17 and the Quick Inventory of Depressive Symptomatology-Self-Report (QIDS-SR16). The HAMD-17 is a clinically used 17-item assessment tool for MDD with a total score ranging from 0 to 52 (Hamilton, Reference Hamilton1960). It has good reliability and validity among Chinese patients with MDD (Zheng et al., Reference Zheng, Zhao, Phillips, Liu, Cai, Sun and Huang1988). A HAMD-17 total score of 0–7 indicates no depressive symptoms, while a HAMD-17 total score of 8–16 or 17–23 indicates mild and moderate depressive symptoms, respectively (Ma et al., Reference Ma, Yang, Yang, Kang, Wang, Zhang, Wang, Zong, Wang, Bai, Guo, Yao, Fang and Liu2021). The QIDS-SR16 is a self-report assessment tool for MDD (Liu et al., Reference Liu, Xiang, Wang, Zhu, Ungvari, Kilbourne, Lai, Zhong, Zhang, Zhang, Zou, Xiao, Zhao, Li, Wu, Zhang and Chiu2013; Rush et al., Reference Rush, Trivedi, Ibrahim, Carmody, Arnow, Klein, Markowitz, Ninan, Kornstein, Manber, Thase, Kocsis and Keller2003). A QIDS-SR16 total score of more than 5 indicates mild depressive symptoms, and more than 10 indicates moderate depressive symptoms (Matsumoto et al., Reference Matsumoto, Sakurai, Aoki, Takaesu, Okajima, Tachimori, Murao, Maruki, Tsuboi and Watanabe2024; Trivedi et al., Reference Trivedi, Rush, Wisniewski, Nierenberg, Warden, Ritz, Norquist, Howland, Lebowitz, McGrath, Shores-Wilson, Biggs, Balasubramani, Fava and Team2006). Additionally, in the field visits and unscheduled visits, the severity of mania and hypomania was evaluated by the 33-item Hypomania Checklist (HCL-33), which is a reliable screening tool for BD in Chinese patients with MDD (Feng et al., Reference Feng, Xiang, Huang, Wang, Feng, Tian, Geng, Ungvari, Chiu, Tse and Angst2016; Zhang et al., Reference Zhang, Li, Zhao, Cheung, Ungvari, Wang, Xiang and Angst2022). An HCL-33 total score of more than 15 indicates a high risk of BD in patients with MDD (Feng et al., Reference Feng, Xiang, Huang, Wang, Feng, Tian, Geng, Ungvari, Chiu, Tse and Angst2016).
In the remote visits, the severities of depressive and manic symptoms were evaluated with the PHQ-9 and MDQ, respectively. The PHQ-9 is a widely used self-report tool for evaluating the severity of depressive symptoms among populations (Kroenke, Spitzer, & Williams, Reference Kroenke, Spitzer and Williams2001). The total score of the PHQ-9 ranges from 0 to 27, and a total score of 5 or more, or 10 or more, indicates potential depressive symptoms (Negeri et al., Reference Negeri, Levis, Sun, He, Krishnan, Wu, Bhandari, Neupane, Brehaut, Benedetti and Thombs2021). The MDQ is a fast-screening tool for manic symptoms which showed good sensitivity, specificity and test–retest reliability (Gervasoni et al., Reference Gervasoni, Weber Rouget, Miguez, Dubuis, Bizzini, Gex-Fabry, Bondolfi and Aubry2009; Hirschfeld et al., Reference Hirschfeld, Williams, Spitzer, Calabrese, Flynn, Keck, Lewis, McElroy, Post, Rapport, Russell, Sachs and Zajecka2000; Kubota et al., Reference Kubota, Yokoyama, Irino, Mitamura, Okatsu, Sasagawa, Kanie, Nakajima, Makino, Aoyama, Kataoka, Horikoshi, Kuga and Ito2026); the MDQ’s total score of 7 or more could be used for screening of BD in psychiatric outpatient (Hirschfeld et al., Reference Hirschfeld, Williams, Spitzer, Calabrese, Flynn, Keck, Lewis, McElroy, Post, Rapport, Russell, Sachs and Zajecka2000).
Covariates
Some baseline demographic characteristics, including sex (men or women), education level (middle school and below, high school, college and university, graduate school and above), marital status (married or unmarried), employment status (yes or no), medical insurance (yes or no), residence (urban or rural), onset age of the first episode of MDD (≤25 or >25 years old), whether this was the first episode of MDD (yes or no), suicide attempts in the current episode (yes or no), psychiatric symptoms in the current episode (yes or no), and family history of mental disorders (yes or no), taking antipsychotics or mood stabilizer (yes or no) were selected as covariates in this study.
Sample size calculation
PASS 15.0 software was used for the sample size determination. Based on the Cox regression model, we determined that 1,730 participants would be required to detect an HR of TTR or stability of symptoms in 0.5 with 90% power at α = 0.05 (two-tailed), assuming a 3% rate of conversion to BD in the patients initially diagnosed with MDD.
Statistical analysis
The continuous variables were presented as mean ± standard deviation (SD) in normal distribution or median (interquartile range) in skewed distribution. The categorical variables were presented as frequencies and percentages (%). The chi-square test was used to compare the distributions of categorical variables in patients with MDD who converted to BD and who did not.
In this study, TTR was calculated as the number of evaluations with the total score being under the predefined cutoff within a window divided by the total number of evaluations within that window. For the field visits, an 8-week landmark (baseline, weeks 2 and 8, and 3 visits) was used to calculate the 8-week TTR (TTR8w) for the HAMD-17, QIDS-SR16, and HCL-33 scores, which were then applied to predict the conversion to BD between Weeks 8 and 52. For the remote visits, a landmark of Week 22 (with biweekly assessments from Weeks 14 to 22, totaling five visits) was used to calculate the TTR14-22w for the PHQ-9 and MDQ scores, predicting conversion between Weeks 22 and 52. Based on previous studies, the cutoff values were set to 7 and 17 for the HAMD-17, 7 and 15 for the HCL-33, 5 and 10 for the QIDS-SR16, 5 and 10 for the PHQ-9, and 7 for the MDQ. The selected cutoffs were either clinically meaningful cutoffs or optimal cutoff points for the scales. The different cutoff values for the same scale were used as sensitivity analyses.
Furthermore, based on the concept derived from the SPC, we used the person-specific stability of symptoms, evaluated by the proportion of time the symptoms stay within the normal ranges that were individually calculated and defined for each person, to quantify the effects of the fluctuation of symptoms on the conversion to BD. First, records with missing visit dates were excluded. Then, participants with fewer than two valid assessments during the calculation window were removed, as the interpolation procedure requires at least two time points to estimate the trajectory. Next, we applied piecewise linear interpolation using the “METHOD = JOIN and OBSERVED=BEGINNING” options in SAS “PROC EXPAND” to fill in missing weekly values, thereby generating a complete weekly time series from baseline to the last visit for each participant. Then, the 25th and 75th percentiles were calculated for each participant’s time series. Scores between the 25th and 75th percentiles were considered within the individual’s ‘normal’ range. Scores below the 25th and above the 75th percentiles were considered outside the ‘normal’ range. The SPC-evaluated person-specific stability (hereinafter referred to as ‘SPC’) was subsequently computed as the number of time points during which the participant’s score fell within their normal range, divided by the total number of time points within the respective time window. Following the same approach as TTR, SPC8w was calculated using the landmark of Week 8 (baseline to Week 8, nine visits) for the HAMD-17, QIDS-SR16, and HCL-33 scores in field visits. Similarly, SPC14-22w was calculated using the landmark of Week 22 (Week 14–22, nine visits) for the PHQ-9 and MDQ scores.
Cox regression models were used to evaluate the associations between the TTR and the SPC with the risk of conversion to BD by hazard ratios (HRs) and 95% confidence intervals (CIs). The proportional hazards assumption for each index was checked. The Spearman correlation analysis was used to examine the colinearity among the covariates. Three models were applied: Model 1 was the crude model that only included the TTR or SPC in the independent variables; Model 2 was adjusted for sex, onset age of the first episode of MDD, family history of mental disorders, suicide attempts in the current episode of MDD, psychiatric symptoms in the current episode of MDD, first episode of MDD (no or yes), taking psychotics or mood stabilizer (no or yes), and baseline HAMD-17 total scores; Model 3 was adjusted for the variables in Model 2 plus the socioeconomic variables, including marital status, employment status, and medical insurance. Furthermore, in a sensitivity analysis, the patients whose PHQ-9 evaluation was elicited solely by the depressive-symptom criterion (PHQ-2, i.e. the sum of Items 3 and 4, ≥ 3) were included. Additionally, to test the robustness of our findings against the choice of the landmark time point, we performed a sensitivity analysis using the full 52-week observational data, in which the TTR and SPC were calculated over the entire 52-week period and their associations with conversion were assessed using the same Cox proportional hazards model. Subgroup analyses were performed to stratify the associations of the TTR and SPC with the conversion to BD by sex, first episode of MDD, onset age of the first episode of MDD, psychiatric symptoms in the current episode of MDD, and family history of mental disorders. All statistical analyses were performed with the SAS software version 9.4 (SAS Institute). A two-sided P value <0.05 was considered statistically significant.
Results
Baseline characteristics
A total of 2,640 patients were enrolled from 20 clinical centers in 13 provinces across China from May 12, 2022 to February 8, 2025. Of these, 28 patients were excluded due to baseline dropout, 219 were excluded because the follow-up duration exceeded 400 days (in other words, more than 52 weeks), and 26 patients were excluded because their outcome could not be determined due to the absence of records of the exact dates of the relevant follow-up visits. Therefore, 2,367 patients were finally included in the analysis, and 69 patients converted to BD during the 52-week follow-up, with a conversion rate of 2.92% (2.23%–3.59%). The flowchart of participant recruitment can be found in Supplementary Figure S1. Of them, 61 patients converted to BD between Weeks 8 and 52, and 31 converted to BD between Weeks 22 and 52. The mean age of the 2,367 patients was 29.88 ± 9.79 years; 33.46% (n = 792) of the patients were men. The median follow-up time for patients converted to BD was 137 days, ranging from 13 to 353 days. The number of events during the follow-up could be found in Supplementary Table S1. The chi-square tests or Fisher’s exact tests showed that patients who were unmarried, did not have medical insurance, and were currently working had statistically higher possibilities of converting to BD (P < 0.05). Sex, onset age of the first episode of MDD, relapse, suicide attempts in the current episode of MDD, psychiatric symptoms in the current episode of MDD, family history of mental disorders, education level, and urban–rural residence were not statistically different between patients with MDD who converted to BD and who did not (P < 0.05) (Table 1). Figure 2 presents the average symptom fluctuations over time after interpolation. Patients who converted to BD during the follow-up period exhibited greater fluctuations in both depressive and manic symptoms than those who did not convert.
Baseline characteristics of participants
Table 1. Long description
The table consists of six columns: Variables, Total (n = 2,367), Not conversion to BD (n = 2,298), Conversion to BD (n = 69), chi-squared, and P value.
Key findings include:
* Sex: 792 Men (33.46 percent) and 1,575 Women (66.54 percent). P value is 0.424.
* First episode of MDD: 1,395 Yes (58.99 percent) and 970 No (41.01 percent). P value is 0.747.
* Suicide attempts in this episode: 2,246 No (94.97 percent) and 119 Yes (5.03 percent). P value is 0.393.
* Psychiatric symptoms: 2,291 No (96.87 percent) and 74 Yes (3.13 percent). P value is 0.475.
* Family history of mental disorders: 1,967 No (83.21 percent) and 397 Yes (16.79 percent). P value is 0.149.
* Education level: Majority are College and university (1,525 participants, 65 percent). P value is 0.547.
* Residence: 339 Rural (14.68 percent) and 1,970 Urban (85.32 percent). P value is 0.220.
* Marital status: 761 Married (32.42 percent) and 1,586 Unmarried (67.58 percent). P value is 0.029, indicating statistical significance.
* Medical insurance: 2,047 Yes (90.18 percent) and 223 No (9.82 percent). P value is 0.005, indicating statistical significance.
* Work status: 1,253 Yes (53.66 percent) and 1,082 No (46.34 percent). P value is 0.019, indicating statistical significance.
* Onset age: 2,032 were less than or equal to 25 years old (85.85 percent). P value is 0.788.
* Antipsychotics or mood stabilizer: 47 Yes (1.99 percent) and 2,320 No (98.01 percent). P value is 0.646.
Note: BD stands for bipolar disorder; MDD stands for major depressive disorder.
Note: BD, bipolar disorder; MDD, major depressive disorder.
a The Fisher’s exact test.
The average fluctuations of symptoms along with time after interpolation. (A) the 17-item Hamilton Depression Scale (HAMD-17); (B) the 33-item Hypomania Checklist (HCL-33); (C) the 16-item Quick Inventory of Depressive Symptomatology-Self-ReportMood (QIDS-SR16); (D) the Mood Disorder Questionnaire (MDQ); (E) the 9-item Patient Health Questionnaire (PHQ-9). Status = 0 represents patients did not have the diagnostic conversion to bipolar disorder; Status = 1 represents patients had the diagnostic conversion to bipolar disorder during the follow-up.
Figure 2. Long description
A multi-panel figure containing five line graphs labeled A through E. Each graph plots ‘time (week)’ on the x-axis from 0 to 150 and ‘Average total score’ on the y-axis. A legend in each panel identifies blue lines as status 0 (no bipolar conversion) and brown lines as status 1 (bipolar conversion).
* Panel A (H AMD-17): The blue line starts near 25, drops sharply, and stabilizes around 10. The brown line starts later at week 15 near 30, fluctuates significantly, and generally remains above the blue line until week 140. Horizontal dashed lines are at scores 7 and 17.
* Panel B (H CL-33): The blue line remains stable near 10. The brown line starts at week 15 near 23, peaks at 25 around week 45, and fluctuates between 15 and 20. Horizontal dashed lines are at 7 and 15.
* Panel C (Q IDS-SR16): The blue line starts at 20 and settles near 10. The brown line starts at week 15 near 20, fluctuates, and stays slightly above the blue line. Horizontal dashed lines are at 5 and 10.
* Panel D (M DQ): The blue line is stable between 2.5 and 5. The brown line starts at week 30 near 6, fluctuates, and shows a sharp upward spike toward 10 at week 145. A horizontal dashed line is at 7.
* Panel E (P HQ-9): Both lines start around week 20-30. The blue line fluctuates between 10 and 15. The brown line shows high volatility, including a sharp drop to 5 at week 70 before returning to the 10-15 range. Horizontal dashed lines are at 5 and 10.
Associations between the TTR of symptoms and the conversion to BD
Regarding depressive symptoms, the TTR of the scores on the HAMD-17 (field visits), QIDS-SR16 (field visits), and PHQ-9 (remote visits) were calculated with different thresholds. Cox regressions revealed that the TTR8w of the QIDS-SR16 scores under the cutoff value of 10 was significantly negatively associated with the risk of conversion, indicating that the higher the proportion of time depressive symptoms stayed below the thresholds, the lower the probabilities of converting to a BD diagnosis (Table 2). In addition, the TTR of manic symptoms was calculated with the HCL-33 (field visits) and MDQ (remote visits) scores. Similarly, the Cox regression models suggested that the TTR8w of the HCL-33 scores under the cutoff values of 7 or 15 were negatively associated with the diagnostic conversion to BD (Table 2). Three models showed consistent results after controlling for clinical covariates or social demographic covariates. Subgroup analysis revealed that the TTR8W of the HCL-33 was significantly negatively associated with the conversion to BD in patients with an early onset age of first episode of MDD (≤25 years old) and women (Supplementary Tables S2–S5, Supplementary Figures S2–S8).
The Cox regression models of the associations between TTR and conversion to BD for scales in different cutoff values
Table 2. Long description
The table consists of nine columns: Variables, Total n, Event n, and three models (Model 1, Model 2, and Model 3), each with sub-columns for H R (95 percent C I) and P value.
* HAMD-17 (T TR sub 8w):
- Cutoff 7: Total 2,367, Event 61. Model 1 HR 0.184, P 0.045. Model 2 HR 0.282, P 0.139. Model 3 HR 0.332, P 0.197.
- Cutoff 17: Total 2,367, Event 61. Model 1 HR 0.377, P 0.014. Model 2 HR 0.688, P 0.430. Model 3 HR 1.008, P 0.988.
* HCL-33 (T TR sub 8w):
- Cutoff 7: Total 2,333, Event 61. Model 1 HR 0.249, P 0.001. Model 2 HR 0.258, P 0.001. Model 3 HR 0.231, P 0.001.
- Cutoff 15: Total 2,333, Event 61. Model 1 HR 0.321, P less than 0.001. Model 2 HR 0.321, P less than 0.001. Model 3 HR 0.332, P less than 0.001.
* QIDS-SR16 (T TR sub 8w):
- Cutoff 5: Total 2,357, Event 61. Model 1 HR 0.358, P 0.150. Model 2 HR 0.502, P 0.341. Model 3 HR 0.671, P 0.586.
- Cutoff 10: Total 2,357, Event 61. Model 1 HR 0.349, P less than 0.001. Model 2 HR 0.348, P less than 0.001. Model 3 HR 0.351, P 0.001.
* MDQ (T TR sub 14 to 22w):
- Cutoff 7: Total 825, Event 26. Model 1 HR 0.621, P 0.359. Model 2 HR 0.591, P 0.318. Model 3 HR 0.503, P 0.199.
* PHQ-9 (T TR sub 14 to 22w):
- Cutoff 5: Total 783, Event 20. Model 1 HR 0.320, P 0.479. Model 2 HR 0.445, P 0.621. Model 3 HR 0.525, P 0.686.
- Cutoff 10: Total 783, Event 20. Model 1 HR 0.424, P 0.186. Model 2 HR 0.503, P 0.303. Model 3 HR 0.479, P 0.282.
Note: HAMD-17, 17-item Hamilton Depression Rating Scale; QIDS-SR16, 16-item Quick Inventory of Depressive Symptomatology-Self-Report; HCL-33, 33-item Hypomania Checklist; MDQ, Mood Disorder Questionnaire; PHQ-9, 9-item Patient Health Questionnaire; Model 1, no covariates; Model 2, adjusted for sex, onset age of MDD, family history of mental disorders, having psychiatric symptoms in this episode, having suicide attempts in this episode, whether this is the first episode of MDD, baseline HAMD-17 total scores and taking antipsychotics or mood stabilizer; Model 3, additional adjusted for work, medical insurance, marriage based on model 2. TTR, time in target range; HR, hazard ratio; CI, confidence interval.
Associations between the SPC and the conversion to BD
Cox regression models revealed that within 8 weeks, only the SPC8w of the HCL-33 scores was significantly negatively associated with conversion, whether controlling for covariates or not (Table 3). The subgroup analyses of SPC also showed that, the proportions of HCL-33 scores in normal range were significantly negatively associated with the conversion to BD in the group with an onset age of ≤25 years old (HR = 0.028, 95% CI: 0.003–0.241, P = 0.001) and for women (HR = 0.013, 95% CI: 0.001–0.196, P = 0.002) (Supplementary Tables S6–S9, Supplementary Figures S9–S11). Notably, the predictive effect of SPC14-22w on the conversion to BD may be limited by the lower incidence of events between Weeks 22 and 52, as participants with missing visit dates or fewer than two valid assessments were excluded from the SPC calculation.
The Cox regression models of the associations between the SPC and conversion to BD for scales
Table 3. Long description
The table consists of nine columns: Variables, Total (n), Event (n), and three models (Model 1, Model 2, Model 3), each with sub-columns for HR (95% C I) and P value.
* HAMD-17 (S PC sub 8w): Total 2,158, Event 60. Model 1 HR 0.123 (P 0.093); Model 2 HR 0.122 (P 0.086); Model 3 HR 0.127 (P 0.099).
* QIDS-SR16 (S PC sub 8w): Total 2,155, Event 60. Model 1 HR 0.359 (P 0.305); Model 2 HR 0.291 (P 0.213); Model 3 HR 0.343 (P 0.281).
* HCL-33 (S PC sub 8w): Total 2,144, Event 60. Model 1 HR 0.028 (P 0.001); Model 2 HR 0.018 (P less than 0.001); Model 3 HR 0.019 (P less than 0.001).
* MDQ (S PC sub 14 to 22w): Total 457, Event 15. Model 1 HR 0.176 (P 0.199); Model 2 HR 0.231 (P 0.306); Model 3 HR 0.380 (P 0.514).
* PHQ-9 (S PC sub 14 to 22w): Total 450, Event 12. Model 1 HR 1.566 (P 0.771); Model 2 HR 1.272 (P 0.879); Model 3 HR 0.920 (P 0.961).
Model 1 has no covariates. Model 2 adjusts for sex, onset age, family history of mental disorders, psychiatric symptoms, suicide attempts, whether this is the first episode of MDD, baseline scores and taking antipsychotics or mood stabilizer. Model 3 adds work, insurance, and marital status.
Note: HAMD-17, 17-item Hamilton Depression Rating Scale; QIDS-SR16, 16-item Quick Inventory of Depressive Symptomatology-Self-Report; HCL-33, 33-item Hypomania Checklist; MDQ, Mood Disorder Questionnaire; PHQ-9, 9-item Patient Health Questionnaire; Model 1, no covariates; Model 2, adjusted for sex, onset age of MDD, family history of mental disorders, having psychiatric symptoms in this episode, having suicide attempts in this episode, whether this is the first episode of MDD, baseline HAMD-17 total scores and taking antipsychotics or mood stabilizer; Model 3, additional adjusted for work, medical insurance, marital status based on model 2. SPC, statistical process control; HR, hazard ratio; CI, confidence interval.
Sensitivity analysis
The sensitivity analysis excluding patients whose PHQ-9 evaluation was elicited solely by anxiety symptoms (GAD-2, i.e. the sum of Items 1 and 2, ≥ 3) did not affect the nonsignificant associations between the TTR14-22w of PHQ-9 scores and the conversion to BD (Supplementary Table S10). In addition, the TTR52w and SPC52w were also computed to explore the associations between symptom fluctuations and the conversion to BD within 52 weeks. The results showed that when the calculation window was enlarged, the TTR52w of HAMD-17 scores with a cutoff value of 17, as well as of the QIDS-SR16, HCL-33, and MDQ scores, had significant negative associations with the conversion to BD (Supplementary Table S11). The SPC52w of the HCL-33 and MDQ scores had significant negative associations with the conversion to BD (Supplementary Table S12).
Discussion
The current study analyzed the data at multiple follow-up visits within 1 year of 2,367 patients from the PROUD study. During the 52-week follow-up, 69 patients converted to BD, with a conversion rate of 2.92% (2.23%–3.59%), which is in agreement with previous findings that first-year conversion rates varied from 1.5% (Musliner & Ostergaard, Reference Musliner and Ostergaard2018) to 3.9% (Kessing, Willer, Andersen, & Bukh, Reference Kessing, Willer, Andersen and Bukh2017). After correcting for covariates, it was found that the TTR of depressive symptoms measured with the QIDS-SR16 and manic symptoms evaluated with the HCL-33 were significantly negatively associated with the risk of conversion to BD from MDD. Also, this study found that only the SPC reflected with the HCL-33, but not other scales, was significantly negatively associated with the conversion to BD.
This study found that the TTR of both depressive and manic symptoms of patients with MDD had significant negative associations with the conversion to BD, meaning that the more unstable the symptoms were, no matter depressive or manic, the more likely the patients initially diagnosed with MDD were to have a diagnostic conversion to BD. These findings echoed the well-established understanding that BD is characterized by fluctuations in patients’ mood state (Grande et al., Reference Grande, Berk, Birmaher and Vieta2016). In clinical practice, the findings implied that when treating patients diagnosed with MDD, it may be advisable for clinicians to keep up their vigilance even after a patient’s short-term improvement or remission in terms of depressive symptoms. The close monitoring of both depressive and manic symptoms, even in patients seemingly with depression only, may be beneficial.
Complementarily, while the TTR quantified the fluctuation of symptoms with the proportion of time that relevant scores stayed under the predetermined ‘safe’ thresholds, the SPC offered information on the shifts of the severity of symptoms based on the deviations from individually calculated normal ranges that were unique to each patient, with the potential to provide more timely and personalized feedback. In the current study, it was found that the SPC of symptoms evaluated with the HCL-33 had a significant negative association with the diagnostic conversion to BD. Also, the SPC of the HCL-33 scores had significant negative associations with the conversion to BD in women and in patients with an onset age of ≤25 years. The findings suggested that the severity of manic symptoms should be closely monitored in corresponding patient populations, and a shift of symptoms that exceeds the individually calculated normal range should trigger an alert for clinicians and call for a reevaluation for BD as soon as possible in case of a potential diagnostic conversion.
The concept of Measurement-Based Care (MBC) for MDD was formally defined by Trivedi et al. in the STAR*D study (Trivedi et al., Reference Trivedi, Rush, Wisniewski, Nierenberg, Warden, Ritz, Norquist, Howland, Lebowitz, McGrath, Shores-Wilson, Biggs, Balasubramani, Fava and Team2006). Nowadays, MBC can be defined more generally as the clinical practice of routinely measuring various clinical outcomes with validated tools at visits and using the measurements to guide broader clinical decision-making (Hong et al., Reference Hong, Murphy, Michalak, Chakrabarty, Wang, Parikh, Culpepper, Yatham, Lam and Chen2021). Studies revealed that patients with mood disorders receiving MBC had better outcomes than those receiving standard care (Guo et al., Reference Guo, Xiang, Xiao, Hu, Chiu, Ungvari, Correll, Lai, Feng, Geng, Feng and Wang2015; Zhu et al., Reference Zhu, Hong, Yang, Yang, Wang, Liu, Murphy, Michalak, Wang, Yatham, Chen and Lam2021). For the monitoring of the diagnostic conversion to BD, compared to standard care in which the assessments are administered based on the clinician’s discretion, MBC, in which assessments are routinely and regularly administered and calculated based on the concepts of the TTR and SPC, could provide quantified, timely information on the fluctuation of symptoms. However, only a small proportion of practitioners employed MBC in their daily practice (Zimmerman & McGlinchey, Reference Zimmerman and McGlinchey2008), and the measurement of outcomes is still not routine in psychiatry (Hong et al., Reference Hong, Murphy, Michalak, Chakrabarty, Wang, Parikh, Culpepper, Yatham, Lam and Chen2021). Also, unlike for many physical diseases (such as measuring blood pressure in patients with hypertension), it may be less clear or direct as how to measure the outcomes in psychiatry. The findings of the current study not only supported and emphasized the importance of the application of MBC but also, with statistical evidence, provided more specific guidance on measurement tool selection for detecting conversion in target populations.
The findings suggested that it may be beneficial to incorporate certain scales that are not conventionally included in regular assessments for patients diagnosed with MDD. It was found that in addition to the QIDS-SR16 scores, the fluctuation in symptoms reflected by the HCL-33 scores also had significant negative associations with the diagnostic conversion. As screening tools for manic/hypomanic symptoms, the HCL-33 is most likely to be used as a one-time screening tool to screen for bipolar features in patients with MDD (Feng et al., Reference Feng, Xiang, Huang, Wang, Feng, Tian, Geng, Ungvari, Chiu, Tse and Angst2016), and it is still not common to regularly administer this scale and monitor relevant changes in clinical practice. Our findings highlighted the potential significance of this screening tool in the long-term management of MDD.
Also meaningfully, findings of the current study provided evidence-based clues for building MBC plans for specific purposes and populations. For instance, as long-standing measurement tools, all scales employed in the current study have been proven to have satisfactory psychometric properties. In clinical practice, those scales may be viewed as interchangeable and chosen simply out of availability, convenience, or habits of the practitioner. However, a noteworthy finding was that the changes reflected with the PHQ-9, HAMD-17, or MDQ scores were not statistically associated with the conversion to BD, no matter calculated based on the concepts of the TTR or SPC. While all scales were found generally sensitive to small changes over time (Feng et al., Reference Feng, Huang, Tian, Wang, Hu, Chiu, Ungvari, Kilbourne and Xiang2016), this finding may be explained by the differences in specific items included in the scales. For instance, it was found that although the PHQ-9 was sensitive to changes at the total-score level, its Item 9 (suicidality) was the least sensitive to change (Ayilara et al., Reference Ayilara, White, Dunleavy, Ferro, Araya, Schmider, Seward and Sajobi2026). As having suicide attempt was found to be associated with the MDD-to-BD conversion (Xu et al., Reference Xu, Chen, Hu, Shen, Chen, Tan, Gao, Chen, Chen, Chen, Yuan and Zhang2022), it was possible that the PHQ-9 was less sufficient in detecting those changes that were more likely to be associated with the risk of conversion. This finding reminded us that clinical decision-making in specific contexts can be much more complicated than simply checking psychometric properties of measurement tools. Specifically, while the PHQ-9 is often favored for its brevity and is found to be the most widely used self-rating scale in China (Zhao et al., Reference Zhao, Gui, Xu, Guo, Li, Wang, Wang and Meng2025), the current findings suggested that relying on the PHQ-9 only might miss important information in the context of monitoring for the conversion to BD.
One strength of the current study was its large sample enrolled from multiple sites covering various regions of China. In addition, compared to other studies investigating the diagnostic conversion to BD (Tse et al., Reference Tse, Fok, Yim, Leung and Leung2021), it scheduled more follow-up visits with shorter intervals. While those previous studies provided crucial information on the long-term progression of the disease with longer follow-up periods, the current study was able to capture the fluctuation in symptoms more timely and provided important information on the association between the dynamics of conditions and the conversion with frequent follow-up visits. A limitation was that the follow-up duration was comparatively short. Despite this, it was found that the risk of conversion was the highest in the first year, and the conversion rate showed a quadratic relationship with time, with decreases in subsequent years (Kung et al., Reference Kung, Tsai, Cheng, Tsai, Su, Chen, Bai, Liang and Chen2024), and the current study provided meaningful information on the influence of symptom fluctuation in the critical period of 1 year. Another limitation was that the current study only included adult samples in China, future validation in other populations may be informative. It should also be noted that, although socioeconomic status (SES) variables have been adjusted for in this study, as information on most SES variables (for instance, marital status, employment status, etc.) were collected from self-report of the patients instead of official records, it is possible that reporting bias or social desirability bias exists and the reporting of those variables may be inaccurate. Moreover, a possible limitation was that there might be doubt about the criteria for the elicitation of further PHQ-9 administration: as the PHQ-9 administration could be elicited by either anxiety symptoms or depressive symptoms while the two groups may be somewhat heterogeneous, the analysis involving the PHQ-9 scores may be influenced by who were included in the sample. To resolve this doubt, a sensitivity analysis was performed and included only the patients whose PHQ-9 evaluation was elicited solely by the depressive-symptom criterion. In this restricted sample, the PHQ-9-based TTR14-22w was recalculated and yielded results that were highly consistent with the primary analysis. However, owing to the reduction in sample size and the consequently limited number of conversion events, the SPC14-22w could not be reliably estimated and meaningful subgroup interaction analyses could not be conducted, as the models would have been severely underpowered. Similarly, regarding the analysis of the MDQ scores, the elicitation of further MDQ administration could rely on a nonspecific ‘positive mood’ item, it is possible that under this criterion, some patients with normative, stable positive affect instead of true hypomanic/manic liability would be included in the analysis, which might confound the observed association between greater MDQ score stability and lower risk of diagnostic conversion to BD. And for the SPC, the HCL-33 and MDQ scores were treated as continuous variables in a longitudinal framework, while the SPC was more of a supplement to the TTR and was an exploration of the potential of broadening the application of the MDQ and HCL-33, there may still be a certain risk of diluting the construct validity of the scales. Therefore, the results regarding the SPC of the PHQ-9, HCL-33, and MDQ scores should be interpreted with caution. Additionally, the use of linear interpolation between visits assumes linear changes over time, which may smooth out acute fluctuations in symptoms and thus potentially obscure early warning signals preceding conversion. Finally, the current study focused only on the fluctuation of mood state; whether the fluctuation of symptoms in other dimensions (for instance, cognitive function) could also predict the conversion to BD is worth exploring in the future.
Conclusion
The current study found that greater stability of depressive and manic symptoms was associated with a lower risk of conversion to BD in patients with MDD over 52 weeks. Notably, for the HCL-33, a negative association with conversion risk was consistently observed across its person-specific stability metric (SPC) and time-in-range stability metric (TTR) defined using two prespecified cutoffs (≤7 and ≤15) during the first 8 weeks. These findings suggest that although further validation in independent samples is still needed, continuous monitoring of intraindividual variability in hypomanic symptoms may be of clinical significance for the early identification of conversion to BD.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106400.
Data availability statement
The data generated during the study are not publicly available due to data security and privacy but are available from the corresponding author on reasonable request.
Acknowledgements
We would like to thank the doctors and nurses at these research centers: 1. Beijing Anding Hospital, Capital Medical University, 2. West China Hospital of Sichuan University, 3. The Second Xiangya Hospital of Central South University, 4. Tongji Hospital Affiliated to TongjiUniversity, 5. The Sixth People’s Hospital of Hebei, 6. Tianjin Anding Hospital, 7. Shandong Daizhuang Hospital, 8. Nanfang Hospital, Southern Medical University, 9. The First Affiliated Hospital of Kunming Medical University, 10. Sir Run Run Shaw Hospital Affiliated to Zhejiang University School of Medicine, 11. The First Affiliated Hospital, Zhejiang University School of Medicine, 12. Anhui Mental Health Center (The Fourth People’s Hospital of Hefei), 13. Hangzhou Seventh People’s Hospital, 14. The Third Hospital of Mianyang, 15. The Fourth People’s Hospital of Chengdu, 16. Henan Provincial Psychiatric Hospital (The Second Affiliated Hospital of Xinxiang Medical University), 17. Shenzhen Kangning Hospital, 18. Xi’an Mental Health Center, 19. Chaohu Hospital of Anhui Medical University, and 20. The First Hospital of Hebei Medical University. We want to express our gratitude to the doctors and nurses for their support throughout the study and to all the patients for their participation.
Author contribution
H.Q., J.Z., and G.W. designed the study. Z.F., Y.F., L.X., R.L., X.L., J.L., and X.Z. collected the data. H.Q. analyzed the data. H.Q. and Z.F. drafted the manuscript. J.Z. and G.W. revised the intellectual content. All authors participated in the final approval of the article.
Funding statement
This work was supported by the Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2021ZD0200600).
Competing interests
The authors declare that they have no competing interests.
Ethical standards
The ethical approval of this study was obtained from the Beijing Anding Hospital, Capital Medical University, Beijing, China (Approval Number 14. [2022] keyan-20221FS-2). All participants have given written informed consents.
Open access
