Diagnostic conversion from unipolar depression to bipolar disorder, schizophrenia, or schizoaffective disorder: A nationwide prospective 15‐year register study on 43 495 inpatients

To examine temporal patterns and predictors for diagnostic conversion from unipolar depression (UD) to bipolar disorder (BD), schizophrenia, and schizoaffective disorder (SAD).


| INTRODUC TI ON
Unipolar depression (UD), bipolar disorder (BD), and schizophrenia, prevalent major psychiatric disorders, [1][2][3][4] are leading contributors to functional and vocational disability worldwide 5 and to deaths by suicide. 6 Although early diagnosis and treatment of these disorders would significantly reduce their enormous burden, apparent phenomenological similarity in the prodromal phases and clinicians' limited ability to predict the development of psychopathology on the individual level often challenge recognition of the major psychiatric disorders in clinical practice. Depressive symptoms and syndromes are common at the initial stages of both BD [7][8][9] and psychotic disorders. 10,11 As BD and schizophrenia differ in their prognosis and treatment approaches, knowing how commonly patients initially diagnosed with UD will subsequently convert to BD and psychotic disorders, when the diagnostic conversion will commonly occur, and what factors predict the conversion are critical for identifying depressed patients at risk.
The meta-analysis of Kessing et al showed that in 11 studies using survival analysis, the 10-year cumulative risk for conversion from UD to BD was 12.9%, with the highest conversion risk emerging during the first few years of follow-up. 12 A comparable finding arose from the meta-analysis of Ratheesh et al 13 However, both authors emphasized high variation in the finding of conversion risk across studies. Among factors underlying this divergence, several are of particular importance when examining the diagnostic conversion incidence and its temporal patterns. First, as the conversion from UD to BD is more likely to occur within 5 years of the first episode of depression, studies following depressed patients from their first episodes probably report more accurate cumulative conversion incidence than studies on patients with multiple recurrent episodes of UD. Second, retrospective study design commonly yields a higher conversion risk than prospective design, possibly due to a risk of recall or selection bias. Third, type of outcome and setting are crucial. As mania often requires hospitalization, studies including only inpatients may underestimate the conversion to BD type 2 since the diagnostic conversion occurring in outpatient care remains undetected. Finally, although clinical cohort studies produce diagnostically more accurate data, register-based studies may produce representative national-level generalizable data. Thus, when characterizing the temporal patterns of the diagnostic conversion, a prospective design, a large sample size, and a long follow-up of patients with UD from their initial phase of illness are vital.
Evidence on predictors of the conversion from UD to BD is also inconsistent. 12,13 The meta-analysis of Kessing et al found no risk factors that were consistently confirmed across studies. 12 Factors comparable to those influencing divergence on the conversion risk from UD to BD across studies are likely to also explain varying findings on the risk factors. 12 While prospective clinical cohort studies may examine the predictive values of broad sets of risk factors, 14 large-scale register-based studies can commonly investigate only a limited number of predictors.
Moreover, as longitudinal studies following patients from their first depressive episode estimate conversion risk more accurately, it is likely that these studies also estimate more accurately the predictive values of risk factors for it. 12 Consequently, prospective studies with a large sample size, following the patients from their first depressive episodes for a long time are needed to estimate the predictive values of the risk factors. To our knowledge, no longitudinal register-based study has explored potential secular trends in risk of conversion. The reasons for examining them include deinstitutionalization and reductions in available hospital beds, and abundant epidemiological evidence for under-recognition of BD, also from Finland. 15 Improvements in diagnostic procedures could significantly increase the rates of conversion over time. The use of antidepressants has been associated with emergence of hypomania or mania, 16 and has also markedly increased in Finland between 1990 and 2010. 17 Therefore, comparing rates of conversion between time periods is warranted.
Depressive symptoms and UD frequently precede both schizophrenia 11,18,19 and schizoaffective disorder (SAD), 20 and ultra-highrisk (UHR) patients often seek treatment due to depressed mood. UD per se appears to predict conversion to psychosis in UHR patients. 10 The Danish prospective register-based study on transition from UD to schizophrenia showed that the 18.5-year cumulative incidence of conversion from UD to schizophrenia was more than 8% in men and almost 6% in women. 21 As in conversion to BD, the highest conversion to schizophrenia occurred within the first years of follow-up. Younger age, greater depression severity, inpatient treatment, and psychotic depression were among the predictors for conversion from UD to schizophrenia. Overall, these risk factors appear partially overlapping with the predictors for conversion to BD. To our knowledge, the findings by Musliner et al have not been replicated in nationwide register studies, and no register-based studies on conversion to SAD exist.
The aim of this Finnish nationwide prospective register-based study on patients with first hospitalization due to UD is (a) to investigate cumulative incidence of diagnostic conversion from UD to BD, schizophrenia, or SAD over a period of up to 15 years, (b) to characterize the temporal pattern of diagnostic conversion, and (c) to examine five putative predictors of diagnostic conversion. In addition, we examined whether the hazard for conversion differed between time periods 1996-2003 and 2004-2011.

| Study design and data
This study uses data from the MERTTU research project, the background and methodology of which have been reported in detail elsewhere. [22][23][24] In our study, the data are based on the Finnish Hospital FHDR includes data on, for example, admission and discharge dates, diagnoses, operations, hospital, and specialty; quality of data in the register is known to be good. 25 Since 1994, also information on psychiatric admissions, including data on treatments and admission as well as status of overall functioning, has been collected. The overall functioning is assessed by the attending doctor using the Global Assessment Scale (GAS). 26 The Finnish personal identity codes enabled us to connect these registers at the individual level, resulting in a comprehensive dataset on all inpatient psychiatric hospitalizations from 1980 to 2011. No private psychiatric hospitals exist in Finland.
No data on outpatient settings were available for the current study.  Figure 1).

| Outcome
The outcomes were defined by the diagnoses of hypomania or mania (F30.x), BD (F31.x), schizophrenia (F20.x), or SAD (F25.x) according to the ICD-10. We used a minimum time period of 2 weeks between the discharge from the first hospitalization due to UD, to the admission of the following hospitalization with diagnostic conversion registered at discharge. Subsequent hospitalizations with time interval shorter than 14 days were treated as one hospitalization. The followup was ceased when one (the first) of these outcome diagnoses occurred. Due to high heterogeneity among patients with SAD, we also used a dichotomous outcome in the sensitivity analyses: BD (including patients with a manic and mixed type of SAD) and schizophrenia (including patients with a depressive type of SAD).

| Predictors
We investigated the effect of the following predictors on the conversion risk from UD to the major psychiatric disorders: sex, first (single) depressive episode vs recurrent depression, severity of depression (mild, moderate, severe without psychotic symptoms, severe with psychotic symptoms), age at first depressive episode, presence of personality disorder, GAS score at first hospitalization due to UD

| Statistical analysis
Data preprocessing was carried out in SAS 9.4 and statistical analy- PaƟents with their first-ever psychiatric inpaƟent admission due to UD since 1996 from the registers (n = 5 4975) PaƟents with earlier psychiatric inpaƟent care (within 16 years) before the UD admission (n = 31,606) Final study populaƟon (n = 43,495) Died before discharge home (n = 82) PaƟents aged less than 13 years or more than 60 years (n = 9 031) Had a comorbid neurodegeneraƟve disorder, schizophrenia or other non-affecƟve diagnosis, or manic/mixed affecƟve diagnosis (n = 2 397) current situation better, and there were no essential differences. 28,29 We checked the proportional hazards assumption using Schoenfeld residuals and concluded that subdistribution hazard ratios (SHRs) should be interpreted as the average effects over time.

| Sample characteristics
The clinical characteristics of the patients are presented in Table 1.
The mean age was 34.4 (SD 13.9) years, and the majority of patients were females (59%). Almost 85% of the patients received a diagnosis of single depression at their first hospitalization, and the rest had probably been diagnosed with depression earlier in outpatient care.
Of depression diagnoses, 13% were characterized as psychotic, 35% as severe without psychotic symptoms, and the remainder as mild (4%), moderate (28%), or not otherwise specified (20%). More than 70% of patients were assessed as having a severe functional impairment (GAS < 47) on admission.

| Conversions to major psychiatric disorders
The sample was followed for a total of 294 992 person-years. The mean follow-up was 6.78 years, the median 6.37 years. Overall, the cumulative conversion incidence during a 15-year follow-up of patients with UD to other major psychiatric disorders was 11.1% (95% CI 10.7-11.6). The overall cumulative incidences for all major psychiatric disorders are shown in Figure 2.

| Conversion to BD
The cumulative incidence of conversion from UD to BD was  Figure S1 for details). The highest yearly crude incidence rate for BD

| Conversion to major psychiatric disorders for patients with first hospitalization due to psychotic depression
The overall cumulative incidence for conversion to all major psy-  Figure 4).

| Predictors of conversion to major psychiatric disorders
SHRs for predictors of conversion from UD to the major psychiatric disorders are presented in Table 2. Overall, females were at higher risk to convert from UD to BD or SAD than males, whereas males were at higher risk to progress to schizophrenia than fe-  We have also conducted sensitivity analyses with a multivariable Cox regression using different time periods from 2 weeks to 6 months between the first hospitalization due to UD and occurrence of the diagnostic conversion. All cause-specific hazard ratios of the predictors remained within the confidence intervals demonstrated in our sample (with the original definition of the minimum of 2 weeks interval between the first hospitalization due to UD and the diagnostic conversion). The results are available on request.

| Sensitivity analysis for dichotomous outcome
We also conducted the sensitivity analysis using a dichotomous outcome. The cumulative conversion incidence for BD (including patients with ICD-10 diagnosis of a manic or mixed type of SAD) was 7.7% (95% CI 7.4-8.1) and for schizophrenia (including patients with ICD-10 diagnosis of depressive type of SAD) 3.4% (95% CI 3.2-3.7).
In terms of predictors, n0 significant differences between the findings for the three outcomes emerged (see Table S1).

F I G U R E 3
Yearly crude incidence rate for the conversion from unipolar depression (UD) to bipolar disorder (BD), schizophrenia, and schizoaffective disorder (SAD) during a 15-y follow-up of inpatients after their first hospitalization due to UD (n = 43 495)

F I G U R E 4
Cumulative conversion incidence to bipolar disorder, schizophrenia, and schizoaffective disorder during a 15-y follow-up of inpatients with first hospitalization due to psychotic depression (n = 5517)

| D ISCUSS I ON
In this prospective nationwide register study, which covered all first psychiatric hospitalizations due to UD in Finland (n = 43 495) between 1996 and 2011, the overall 15-year cumulative incidence of conversion to another major psychiatric disorders was 11.1%.
More specifically, the 15-year cumulative incidences of conversion to BD, schizophrenia, and SAD were 7.4%, 2.5%, and 1.3%, respectively. The highest cumulative incidence for all of the major psychiatric disorders occurred during the first years of follow-up, thereafter decreasing with time. The most consistent predictor for conversion to a manic, hypomanic, or mixed episode of BD, schizophrenia, or SAD was severe depression, particularly psychotic depression. In patients hospitalized due to psychotic depression, the cumulative risk for conversion to any of the major psychiatric disorders was almost twofold higher than in patients with mild depression. Risk for progression to schizophrenia and SAD was several times higher than in patients with mild depression. Patients with recurrent episodes of depression are more likely than those with a single episode to progress to BD. Females were at higher risk of converting to BD, whereas males had a higher risk of converting to psychotic disorders. Greater psychiatric disturbance within depression predicted diagnostic conversion to BD. Finally, the risk for conversion to the psychotic disorders TA B L E 2 SHRs for predictors of conversion from unipolar depression to major psychiatric disorders after first hospitalization (n = 43 495)

Schizoaffective disorder SHR (CI)
Gender 1.2 (1.07-1.4)  1.3 (1.1-1.4)  1.3 (1.2-1.4 Our register-based prospective study has several strengths. First, we were able to investigate the cumulative incidence of diagnostic conversion and its temporal patterns and predictors in a large representative national Finnish cohort (n = 43 465) of patients with UD. Therefore, our findings are generalizable within Finland and likely to the other Nordic countries with broadly similar healthcare services. Generalizability to other settings remains to be investigated. Second, our dataset included adolescents in the age group of 13-17 years. As BD type 1 and schizophrenia may manifest before the age of 18 years, 30,31 we were able to investigate the conversion risk in this particularly vulnerable age group.
Third, prospective design of the study is an obvious strength when examining the incidence of diagnostic conversion and the predic- Our study also has limitations. First, our dataset is limited to information available on psychiatric hospitalizations. Thus, diagnostic conversions detected in relation to treatment in outpatient psychiatric, primary care, and private psychiatric settings remain unknown, unless the patients were later rehospitalized. Therefore, our dataset presumably consists of patients with the most severe mood and psychotic disorders, and the less severe disorders are not included. As hypomania, as opposed to mania, rarely requires hospitalization, we assume that our findings reflect the diagnostic conversion to BD type 1 more precisely than the conversion to However, since the highest rate of conversions was found to occur early, the rate of late diagnostic change after 15 years is likely to be low in BD and negligible in schizophrenia and SAD.

| Cumulative incidence and its temporal patterns for diagnostic conversion from UD to BD, schizophrenia, and SAD
In our study, the cumulative incidence of conversion from UD to BD during a 15-year follow-up (7.4%) is lower that reported in the meta-analyses by Kessing et al 12  in Denmark. However, the conversion incidence to schizophrenia was approximately two and half times lower than in the Danish register study. 21 Since lifetime prevalence of schizophrenia in Finland was reported to be even higher than in other Nordic countries, 42,43 this finding is unexpected. We presume that the lack of data on outpatient units and private settings, the register-based design, and the exclusion of patients with previous hospitalizations due to unspecified psychosis may partly explain this discrepancy. In addition, slight differences may exist in the mental health service structures between these countries.
In our study, the risk for transition to psychotic disorder was higher during 1996-2003 than during 2004-2011. This finding accords with the previously reported decline in the incidence of schizophrenia in Finland. 44 The risk for subsequent hospitalization due to a depressive episode of BD was slightly higher during the time period 2004-2011 than 1996-2003, whereas the risk for hospitalization due to a manic, hypomanic, or mixed episode remained the same, as did risk of bipolar conversion overall. Improved recognition of BD type 2 by clinicians at outpatient settings during the later time period may underlie this finding. The low conversion risk for SAD corresponds to a low lifetime prevalence of SAD. 45 Moreover, low diagnostic reliability of SAD in clinical practice may underlie the low conversion risk 34 in our study. The yearly crude incidence rate for all major psychiatric disorders appeared highest in the first year after hospitalization due to UD, thereafter decreasing with time.

| Predictors for diagnostic conversion
Predictors associated with conversion to the major psychiatric disorders in our study corresponded to the predictors reported in the studies of the Danish group on conversion to BD 38,46 and schizophrenia. 21 Severe depression, especially psychotic depression, was the most consistent predictor for conversion to any major psychiatric disorder. In our study, one of five patients with a history of hospitalization due to psychotic depression progressed to a major psychiatric disorder. Compared with patients with mild depression, patients with psychotic depression had approximately a fivefold risk for progression to schizophrenia. The hazard ratio for psychotic depression to progress to SAD was exceptionally high, 10.6 (CI 95% 4.0-28.4). This finding may reflect the previously reported frequent shift from psychotic depression to SAD, 20 although the wide confidence interval indicates uncertainty in the actual size of the risk.
In addition to severity of depression, patients with moderate and severe psychiatric disturbance during UD had an approximately twofold higher risk for conversion to BD than patients with mild disturbance. Furthermore, as in earlier studies, 9 we found patients with recurrent (previously non-hospitalized) depression to beat higher risk of converting to BD than patients with a single episode. As in the study of the Danish group, females were at higher risk of conversion to BD than males. 38 47 Poor recognition of BD in adolescence may be partly explained by overlapping symptoms of BD and comorbid disorders such as attention-deficit hyperactivity disorder. 48 In addition, clinical presentation of UD in adolescence may differ from UD in adults, resulting in misdiagnosis of UD in young people. 49 Therefore, exclusion of all psychiatric hospitalizations before the first inpatient treatment due to UD may underestimate the effect of age on conversion.
In contrast, corresponding to studies in UHR patients, 10,11 patients aged 13-17 years were at higher risk of converting to psychotic disorders than patients aged 40-60 years. This finding might be explained by the fact that depressive symptoms and diagnosed depressive disorders are common in the prodromal phase of psychotic disorders in young people. 50 Thus, conversion to psychotic disorders seems to occur earlier than to BD in patients with their first hospitalization due to UD. Finally, patients with comorbid personality disorder were at higher risk of diagnostic conversion to BD. This probably reflects the association between comorbid personality disorder and increased risk for chronicity, recurrence, and severity of depression during a mood disorder episode. 51 Taken together, although prediction of further development of psychopathology is challenging on the individual level, patients with a history of a hospitalization due to psychotic depression, recurrent episodes, and greater psychiatric disturbance within depression are particularly vulnerable for further conversion from UD to BD or psychotic disorder during the first years after the hospitalization. These findings should be taken into consideration when planning the outpatient follow-up of the patients discharged from the hospitalization due to UD.
To conclude, in Finland, among patients with first psychiatric hospitalization due to UD, approximately one in nine seems to convert to a major psychiatric disorder during a 15-year follow-up. The cumulative conversion incidence for BD was 7.4%, corresponding to the previously reported conversion incidence from UD to BD in Denmark. However, the cumulative incidence of conversion to psychotic disorders was relatively low: 2.5% for schizophrenia and 1.3% for SAD. Patients suffering from psychotic depression represent a vulnerable group for conversion, with a higher risk for psychotic disorders than for BD. The highest risk for conversion emerged within the first years of follow-up. Males were at higher risk than females for progression to schizophrenia, whereas females were at higher risk for conversion to BD and SAD. Greater psychiatric disturbance during UD or presence of a comorbid personality disorder predicted conversion to BD. Patients with recurrent depression were at higher risk for conversion to BD than patients with a single depression episode. Patients aged 13-17 years were at higher risk for progression to psychotic disorders than older patients. In contrast, patients older than 18 years were at higher risk for conversion to BD than younger patients. Finally, during a 15-years follow-up, in patients with their first hospitalization due to UD, the risk for the subsequent hospitalization due to schizophrenia and SAD was somewhat higher during the first half of the time period, whereas for bipolar depression during the latter. Thus, time period of investigation may have some influence on observed hospitalization-based rates of diagnostic conversion.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Due to limitations imposed by research permit and legislation, data not available.