Abstract
Objectives
Treatment‐resistant schizophrenia (TRS) is clinically defined as failure to respond to two antipsychotics of adequate dose and duration. An algorithm (registry TRS) was developed, for identifying patients with TRS in claim datasets from Sweden and the United States.
Methods
Schizophrenia (SZ) patients aged ≥13 years were identified in both datasets and matched to controls. Patients were identified as having TRS by use of the registry TRS or ≥1 prescription for clozapine or use of other published criteria. The algorithm was compared for sensitivity, and patients with and without TRS were compared for psychiatric and hospital burden and Global Assessment of Functioning (GAF) scores. TRS prevalence was not assessed due to lack of clinically validated data to test the specificity of the algorithm.
Results
Swedish registry TRS patients ≤45 years at first SZ diagnosis had significantly lower GAF scores and earlier disease onset than non‐TRS patients. SZ patients with higher psychiatric comorbidity and hospital burden were more likely identified as TRS by all algorithms. The registry algorithm was significantly more sensitive to multiple inpatient stays and all psychiatric comorbidities at identifying TRS.
Conclusion
The registry algorithm appeared more sensitive at identifying patients with TRS, who had greater psychiatric and hospital burden.
Keywords: epidemiology, methodology, schizophrenia
1. INTRODUCTION
Treatment resistance is challenging to identify and is a major clinical issue in managing patients with schizophrenia (SZ). Treatment‐resistant schizophrenia (TRS), clinically defined as failure to respond to two trials of different antipsychotics (APs), one usually an atypical, of adequate dose and duration, affects up to one third of individuals with SZ (Hasan et al., 2012; Lehman et al., 2004; National Institute for Health and Care Excellence [NICE], 2014). Adequate dose is often considered as 400–600 mg chlorpromazine equivalent, and duration varies but is often agreed as a minimum of 6 weeks of therapy (Lally et al., 2016; Leucht et al., 2015).
TRS poses the greatest disability of all mental disorders (Iasevoli et al., 2016; Lehman et al., 2004) as patients experience not only persistent core positive symptoms but also negative and/or cognitive symptoms (Hasan et al., 2012; Howes et al., 2017). This results in individuals suffering from poor functional outcomes, increased risk of unemployment, homelessness, substance abuse, imprisonment, agitation, violent victimization, and suicide (Brekke, Prindle, Bae, & Long, 2001; Iasevoli et al., 2016; Jones & Castle, 2006; Kennedy, Altar, Taylor, Degtiar, & Hornberger, 2014). The healthcare burden and related costs are also increased in TRS compared with treatment‐responsive SZ (non‐TRS), driven by longer and more frequent hospitalizations and social services contacts (Kennedy et al., 2014).
Despite some consensus on a clinical presentation of the disease, the identification of patients with TRS for epidemiologic and health services research is not standardized, particularly in automated healthcare databases. Clozapine, recommended by the clinical guidelines (Hasan et al., 2012; Lehman et al., 2004; NICE, 2014) to be used after two AP failures, is the only approved treatment for TRS worldwide, and therefore, prescriptions of the drug can be used as one means to identify patients with TRS in U.S. or European claims data, where clozapine is not used outside of this indication. However, clozapine initiation is typically delayed in favor of increased dosage of the current AP, switching to other APs, or combination therapy (AP polypharmacy; Howes et al., 2012; Stroup, 2014). Various side effects including metabolic syndrome, seizures, and potentially fatal agranulocytosis, myocarditis, and bowel obstruction plus the required routine monitoring lead to underutilization of clozapine in many countries (Corporation, 2017; Kelly, Freudenreich, Sayer, & Love, 2018; Strassnig & Harvey, 2014). Furthermore, only 30–50% of individuals with TRS experience clinically significant improvement with clozapine use (Kane, Honigfeld, Singer, & Meltzer, 1988; McIlwain, Harrison, Wheeler, & Russell, 2011; Siskind, Siskind, & Kisely, 2017). Regulations around clozapine use vary, and due to this lack of global harmonization, clozapine prescriptions are not reliable as a sole means to identify patients with TRS (Nielsen et al., 2016).
Some studies have attempted to use alternative means to identify patients who have TRS, deriving algorithms on the basis of prescribing and hospital events (Stroup, Gerhard, Crystal, Huang, & Olfson, 2016; Wimberley et al., 2016). Stroup et al. (2016) compared treatment outcomes of clozapine versus other AP use in U.S. Medicaid claims data, in patients with evidence of treatment resistance, defined as at least one mental health‐related hospitalization and two or more APs in the past 365 days. Wimberley et al. (2016), using Danish national registry data, used a treatment‐based proxy for TRS, defined as two different APs of at least 6 weeks' duration followed by a hospitalization within 18 months. However, the limitation of both algorithms is the requirement for hospitalization, which may not serve as a reliable marker for TRS, particularly in countries where clinical practice or access to care may limit the use of inpatient psychiatric services for patients with TRS (Levine & Rabinowitz, 2010; Schennach et al., 2012). Further, when studying associations between TRS and resource use, the hospitalization requirement may lead to an overestimation of events, as the algorithm is conditioned on the resource being measured. A study in the United Kingdom (Lally et al., 2016) utilized an alternative algorithm that required two failures of AP treatment at 400 to 600 mg chlorpromazine equivalence for a minimum of 6 weeks (no upper limit) in a small cohort of around 300 patients. However, this algorithm was not tested in larger datasets in other types of health systems, with longer term follow up, and analysis was limited to adults aged 18 to 65.
In this study we developed and tested a new algorithm based upon the work by Lally et al (2016) to identify patients with TRS in health insurance claims data sets. We assessed the face validity of this novel algorithm by comparing characteristics of the patients it identified with those of patients identified by use of either clozapine or Stroup algorithm defined above. Additionally, patients' clinical and hospital burden characteristics were compared between TRS and non‐TRS patients.
2. METHODS
The primary objective of the study was to identify TRS patients among adolescent and adult patients in retrospective datasets in two different health systems, by applying the new algorithm which follows the clinical guideline definitions for the presence of TRS and controlled for non‐adherence and toxicity challenges (Hasan et al., 2012; Lehman et al., 2004; NICE, 2014). Patients with TRS were also identified by prescriptions for clozapine and, in the United States, by the additional algorithm developed by Stroup et al. (2016). The study was not designed to assess prevalence of TRS, as the algorithm necessarily excludes patients who fail the rigorous exposure criteria but in the real world may be considered as TRS by their physicians. Thus, numbers produced in the results should not be interpreted as actual prevalence estimates.
We tested the algorithms in two datasets between years 2005 and 2015 inclusive: (a) the Swedish national disease registry + PsykosR registry (prescribing) and (b) U.S. Medicaid claims data via the “IBM Watson” platform, as publicly funded Medicaid is the source of health care for many patients with SZ. The use of dual datasets in different health systems allowed us to assess the generalizability of the algorithm by comparing any differences in the populations identified, plus the validity of the definition when applied in health systems with varying treatment paradigms.
2.1. Study criteria—Sweden
The study included all patients with an International Classification of Diseases (ICD)‐10 diagnosis code of F20 at an inpatient stay or specialist outpatient visit in the Swedish national patient register. Only patients aged 13 years and over at the date of the first diagnosis were included. The algorithm identified patients as having TRS (named the “registry TRS” population) if they fulfilled the following criteria:
Failure of at least two adequate trials of two different APs, at least one of which must be an atypical AP. An adequate AP trial was defined as a minimum of 6 weeks of treatment with a dose equivalent of 400‐mg chlorpromazine or higher. Treatment failure was defined as treatment discontinuation without prior dose reduction (>20% from average dose during episode), or initiation of another AP drug, within 15 weeks of the first drug's initiation (either as a new monotherapy or as a dual therapy). TRS qualification occurred at the date of the second failure. Fifteen weeks was used as an upper limit, to define “adequate” exposure. It was agreed that continuous usage beyond 15 weeks would not likely be a TRS case.
Patients were also identified at the date of the first script for clozapine and were labeled as the “clozapine” population at that time point.
2.2. Study criteria—the United States
In the U.S. dataset, ICD‐10 code F20 codes ICD 9CM codes 295.x excluding 295.7 were used to define the study population. The cohort was considered a prevalent population as patients are frequently transferred from commercial insurance plans into Medicaid, where linkage between systems is not possible. Therefore, in the U.S. dataset, patients were identified at the time of their first diagnosis of SZ visible in the database. In the United States, patients had to have continuous coverage and eligibility for Medicaid insurance. The algorithm applied in the Medicaid dataset was virtually identical but utilized the clinically recommended international consensus dosing for APs (Gardner, Murphy, O'Donnell, Centorrino, & Baldessarini, 2010), taken as an average over the exposure period, which allowed for some missing data points in patients' dosing histories. A minimum exposure time of both 5 and 6 weeks was also tested, and it was identified that a larger number of patients could be included if the exposure time was decreased to 5 weeks' minimum. Although some guidance only requires a minimum of 2 weeks' AP duration, in a database, it is not possible to see if patients take the prescribed drug (and thus have at least 2 weeks' exposure), only that it was dispensed. Therefore, a second prescription dispensing was considered imperative as a marker of adequate duration.
A clozapine population was also defined as before. The U.S. study also included testing of the registry TRS algorithm against the algorithm developed by Stroup et al. (2016) where TRS was defined as at least three different APs within 365 days (TRS date was defined as the date of the third AP) plus at least one hospitalization for a mental disorder (first listed ICD‐9‐CM 290–319 and E95x) in the 365 days prior to the TRS qualification date. The medication possession ratio for the AP of at least 0.75 was calculated over the 365 days prior to the TRS qualification date. This population was defined as the “Stroup” population.
Patients who did not fulfill any of the TRS algorithms in Swedish or U.S. data were considered as having SZ that was treatment responsive and were defined as “non‐TRS.”
2.3. Matching
Due to lack of prior claims history in the U.S. Medicaid dataset and the resultant prevalent population, it was decided to match all the patients with TRS to non‐TRS patients to remove any potential discrepancies in time with SZ and time with resistance. For this reason, patients were matched on calendar time of first diagnosis of SZ code visible in the dataset, year of birth, and additionally by gender, as there remains lack of clarity regarding male bias in SZ populations, and gender also impacts the pharmacotherapeutic approach to treating psychosis (Hambrecht, Riecher‐Rossler, Fatkenheuer, Louza, & Hafner, 1994; Lange, Mueller, Leweke, & Bumb, 2017). For every patient with TRS, up to 10 non‐TRS patients were matched. For every patient with TRS who dropped out, all matched patients were removed. If matched patients were censored at any time, they were not replaced.
2.4. Follow up and censoring
The study started in January 1, 2005, in the United States and exactly 1 year later in Sweden. The study period ended in December 31, 2015, in both countries. Patient observation for both studies began at the index date (first diagnosis of SZ) and stopped at the end of the study period or upon death, which ever came first.
In the U.S. cohort, only patients with TRS identified by the registry TRS algorithm or Stroup algorithm were censored at the first prescription for clozapine, that is, they could not switch groups to the clozapine group. This was to enable comparison of population types defined by the two TRS algorithms versus clozapine, as clozapine is only used in approximately 2% of SZ patients and may be allocated to specific disease trajectories (Nielsen et al., 2016; Strassnig & Harvey, 2014). In the U.S. Medicaid dataset, patients were also censored from all groups due to loss of eligibility for continuous insurance coverage, which would impact prescribing.
2.5. Outcomes
The outcomes assessed varied by database, dependent upon data availability. As the Swedish data contained more detailed medical information than U.S. claims data, including Global Assessment of Functioning for Symptoms (GAF‐S) scores, it was possible to use these scores to compare TRS and non‐TRS patients as a test of the validity of the algorithm (Hall, 1995; Jones, Thornicroft, Coffey, & Dunn, 1995). The GAF is a 100‐point rating scale where a higher score indicates higher functioning in social, occupational, and psychological well‐being.
GAF scores were not available in U.S. data, and thus, only hospital visits and psychiatric comorbidity previously reported as associated with TRS (Wimberley et al., 2016) and addiction were evaluated, as they were in the Swedish data. The variables included were therefore as follows: age bands 13–17, 18–24, 25–34, 35–44, 45–54, 55–64, 65–74, and 75+; SZ subtype (at first diagnosis claim) identified by ICD‐9CM codes 295.xx; hospital contact rates, that is, inpatient stay (rate and number of bed days per year—≤30 and ˃30) and outpatient visits; episodic depressive disorders (including bipolar and unipolar depression; ICD 9CM 296.xx; 311); personality disorder (all types—301.xx); drug‐induced mental disorder (292.xx); and suicide and/or self‐harm (ICD9: E95xx). These rates were considered in the 365‐day period prior to the TRS qualification date (or matched date in non‐TRS patients).
2.6. Statistical analysis
Descriptive statistics were performed on patients identified as having TRS in both the Swedish and the U.S. datasets, including variables defined previously. In the Swedish dataset, patients with and without TRS were compared by their mean GAF scores using an unadjusted mixed linear effects regression model, observing score since SZ diagnosis. Due to long retention times, results are given over 10 years. Results report values with 95% confidence limits.
Within‐group comparison was conducted for U.S. Medicaid patients with TRS, separately for each algorithm, comparing rates (events/person years of follow up), using the incidence rate ratio (IRR), to evaluate consistency and validity. Patients with TRS were also compared with non‐TRS patients, stratified by each algorithm, to compare population statistics and burden.
3. RESULTS
3.1. Sweden
In the Swedish registry, 31,195 patients had a diagnosis of SZ, and 6,358 of these were patients with a first SZ diagnosis after 2006 and at least one adequate AP treatment period with 15 weeks' follow up (Figure 1a). Of these patients, 562 (8.8%) had failed at least two treatment trials (adequate dose and duration), including one second‐generation AP, indicating treatment resistance and qualifying them for the registry TRS group. Patients with at least one prescription of clozapine totaled 840 (13.2%). Altogether, 1,193 (18.8%) either had two treatment failures and/or were treated with clozapine and were considered as having TRS.
Figure 1.
(a) Patient selection flowchart in Swedish registry. (b) Patient selection flowchart in Medicaid claims data. TRS: treatment‐resistant schizophrenia
Patients who were included in the registry TRS group were diagnosed with SZ at an average mean age of 33.5 years (32.4 years for clozapine), which was approximately 10 years earlier than for non‐TRS patients (mean 44.1 years), as shown in Table 1. Additionally, patients designated TRS by the registry algorithm or clozapine usage had higher proportions of males (64.1% and 68.3%, respectively) than non‐TRS patients (58.2%). Drug abuse and psychiatric comorbidities (personality disorders, unipolar depression, and self‐harm) were more commonly diagnosed in the registry TRS group than in the non‐TRS group in the year preceding the first ever diagnosis of SZ.
Table 1.
Patient demographics in the Swedish registry database
| Variables | Registry TRS | Clozapine TRS | Non‐TRS |
|---|---|---|---|
| n = 562 | n = 840 | n = 5,165 | |
| Women | 202 (35.9%) | 266 (31.7%) | 2,157 (41.8%) |
| Men | 360 (64.1%) | 574 (68.3%) | 3,008 (58.2%) |
| Age group | |||
| 13–17 | 6 (1.1%) | 23 (2.7%) | 25 (0.5%) |
| 18–24 | 153 (27.2%) | 252 (30%) | 660 (12.8%) |
| 25–34 | 179 (31.9%) | 278 (33.1%) | 1,148 (22.2%) |
| 35–44 | 122 (21.7%) | 145 (17.3%) | 1,000 (19.4%) |
| 45–54 | 60 (10.7%) | 87 (10.4%) | 905 (17.5%) |
| 55–64 | 31 (5.5%) | 35 (4.2%) | 681 (13.2%) |
| 65–74 | 8 (1.4%) | 13 (1.5%) | 427 (8.3%) |
| 75+ | 3 (0.5%) | 7 (0.8%) | 319 (6.2%) |
| Age at first schizophrenia diagnosis—mean (SD) | 33.5 (12.3) | 32.4 (12.6) | 44.1 (17.3) |
| Drug and alcohol use and psychiatric comorbidity diagnoses 365 days prior to first schizophrenia diagnosis | |||
| Alcohol use disorder | 28 (5.0%) | 37 (4.4%) | 198 (3.8%) |
| Drug use related disorder | 100 (17.8%)** | 92 (11.0%)** | 351 (6.8%) |
| Personality disorder | 34 (6.1%)** | 34 (4.1%) | 179 (3.5%) |
| Unipolar depression | 68 (12.1%)* | 97 (11.6%)* | 463 (9.0%) |
| Self‐harm | 3 (0.5%)* | 4 (0.5%)* | 3 (0.06%) |
The GAF scores since diagnosis are shown in Figure 2 for patients identified using the algorithm. Over the 10 years, after the first diagnosis of SZ, patient identified as having TRS had significantly and consistently lower GAF scores than non‐TRS patients. By 10 years, the results drop considerably due to small sample size. The model shows that GAF‐S scores were 4.8 points lower for TRS versus non‐TRS patients (p < 0.0001), but this only occurred among those diagnosed at 45 years of age or younger. In patients who were older at diagnosis, there was no significant difference in GAF‐S scores between TRS and non‐TRS patients. Clozapine patients also had consistently and significantly lower GAF scores for 6 years after the SZ diagnosis, after which time there was some overlap of confidence limits indicating no difference between groups (Figure 3).
Figure 2.
Global Assessment of Functioning for Symptoms (GAF‐S) scores by time since diagnosis and treatment‐resistant schizophrenia (TRS) designation in patients <45 years of age
Figure 3.
Global Assessment of Functioning for Symptoms (GAF‐S) scores by time since diagnosis and lifetime clozapine use in all patients
3.2. The United States
Of the original 338,667 SZ patients identified in U.S. Medicaid claims data, the number of patients with TRS identified by the three algorithms numbered 22,082 (Stroup algorithm), 4,560 (clozapine algorithm), and 6,588 (registry TRS algorithm), the latter cut being shown in Figure 1b. The data in Table 2 indicate that the algorithms produce different age distributions, with the registry TRS algorithm identifying more patients aged 35–44. The majority of patients with TRS were white as opposed to black or Hispanic for all algorithm populations. Alcohol dependence was prevalent; for example, the registry TRS algorithm identified patients with TRS who had alcohol dependence at 33% versus 20% in non‐TRS patients. Most smokers (60%) were identified by the registry TRS algorithm (vs. the other two), and this was compared with 40% in the matched non‐TRS group.
Table 2.
Numbers and proportions of patients with and without TRS in the United States
| Variable | Level | Stroup | Clozapine | Registry TRS | |||
|---|---|---|---|---|---|---|---|
| TRS | Non‐TRS | TRS | Non‐TRS | TRS | Non‐TRS | ||
| Sex | Female | 10,816 (49%) | 108,131 (49%) | 2,045 (45%) | 20,447 (45%) | 3,253 (49%) | 32,530 (49%) |
| Male | 11,266 (51%) | 112,631 (51%) | 2,515 (55%) | 25,148 (55%) | 3,335 (51%) | 33,350 (51%) | |
| Age | 13–17 | 1,343 (6%) | 15,907 (7%) | 291 (6%) | 3,740 (8%) | 287 (4%) | 3,526 (5%) |
| 18–24 | 3,352 (15%) | 33,630 (15%) | 807 (18%) | 8,138 (18%) | 1,048 (16%) | 10,368 (16%) | |
| 25–34 | 4,396 (20%) | 43,964 (20%) | 1,056 (23%) | 10,107 (22%) | 1,494 (23%) | 14,874 (23%) | |
| 35–44 | 4,777 (22%) | 47,413 (21%) | 907 (20%) | 9,273 (20%) | 1,591 (24%) | 15,636 (24%) | |
| 45–54 | 5,161 (23%) | 51,253 (23%) | 952 (21%) | 9,566 (21%) | 1,465 (22%) | 14,543 (22%) | |
| 55–64 | 2,616 (12%) | 24,450 (11%) | 471 (10%) | 4,054 (9%) | 583 (9%) | 5,703 (9%) | |
| 65–74 | 361 (2%) | 3,404 (2%) | 66 (1%) | 613 (1%) | 98 (1%) | 996 (2%) | |
| 75+ | 76 (0.3%) | 741 (0.3%) | 10 (0.2%) | 104 (0.2%) | 22 (0.3%) | 234 (0.4%) | |
| Race | Black | 7,408 (34%) | 97,461 (44%) | 1,251 (27%) | 20,099 (44%) | 2,433 (37%) | 28,535 (43%) |
| Hispanic | 339 (2%) | 3,223 (1%) | 89 (2%) | 706 (2%) | 100 (2%) | 993 (2%) | |
| Other | 1,813 (8%) | 17,766 (8%) | 366 (8%) | 3,683 (8%) | 458 (7%) | 4,971 (8%) | |
| White | 12,522 (57%) | 102,312 (46%) | 2,854 (63%) | 21,107 (46%) | 3,597 (55%) | 31,381 (48%) | |
| Drug and alcohol use in past 365 days | |||||||
| Alcohol dependency | Yes | 5,073 (23%) | 40,977 (19%) | 922 (20%) | 8,880 (19%) | 2,176 (33%) | 12,948 (20%) |
| Alcohol‐induced mental disorder | Yes | 747 (3%) | 5,910 (3%) | 104 (2%) | 1,341 (3%) | 345 (5%) | 1,888 (3%) |
| Drug dependency | Yes | 7,517 (34%) | 62,562 (28%) | 1,292 (28%) | 13,653 (30%) | 3,122 (47%) | 19,706 (30%) |
| Tobacco dependency | Yes | 10,992 (50%) | 90,389 (41%) | 2,016 (44%) | 19,250 (42%) | 3,924 (60%) | 28,554 (43%) |
| Psychiatric comorbidities in last 365 days | |||||||
| Personality disorder | Yes | 3,900 (18%) | 14,995 (7%) | 852 (19%) | 3,358 (7%) | 1,562 (24%) | 4,950 (8%) |
| Episodic mood disorder | Yes | 15,122 (68%) | 104,775 (47%) | 2,807 (62%) | 22,339 (49%) | 4,820 (73%) | 31,838 (48%) |
| Self‐harm | Yes | 602 (3%) | 2,049 (0.9%) | 109 (2%) | 498 (1%) | 224 (3%) | 568 (0.9%) |
| Burden in the last 365 days | |||||||
| Number of inpatients visits (any kind) in the last year | 0 | 14,602 (66%) | 192,342 (87%) | 2,921 (64%) | 39,465 (87%) | 3,971 (60%) | 57,373 (87%) |
| 1 | 4,904 (22%) | 21,752 (10%) | 981 (22%) | 4,639 (10%) | 1,470 (22%) | 6,403 (10%) | |
| 2 | 1,515 (7%) | 4,452 (2%) | 375 (8%) | 966 (2%) | 568 (9%) | 1,375 (2%) | |
| 3+ | 1,061 (5%) | 2,216 (1%) | 283 (6%) | 525 (1%) | 579 (9%) | 729 (1%) | |
| Number of SZ hospitalization days in the last year | 0 days | 14,776 (67%) | 193,195 (88%) | 2,952 (65%) | 39,600 (87%) | 4,042 (61%) | 57,562 (87%) |
| 1–29 days | 6,288 (28%) | 25,253 (11%) | 1,111 (24%) | 5,497 (12%) | 1,979 (30%) | 7,609 (12%) | |
| 30+ days | 1,018 (5%) | 2,314 (1%) | 497 (11%) | 498 (1%) | 567 (9%) | 709 (1%) | |
| Number of outpatients visits (any kind) in the last year | 0 | 4,632 (21%) | 83,583 (38%) | 790 (17%) | 17,326 (38%) | 1,561 (24%) | 25,156 (38%) |
| 1 | 4,157 (19%) | 34,540 (16%) | 1,207 (26%) | 7,157 (16%) | 968 (15%) | 9,664 (15%) | |
| 2 | 1,648 (7%) | 14,697 (7%) | 334 (7%) | 2,998 (7%) | 523 (8%) | 4,287 (7%) | |
| 3+ | 11,645 (53%) | 87,942 (40%) | 2,229 (49%) | 18,114 (40%) | 3,536 (54%) | 26,773 (41%) | |
Note. SZ: schizophrenia; TRS: treatment‐resistant schizophrenia.
The registry TRS algorithm appears more sensitive at identifying patients with psychiatric comorbidities where claims were submitted in the 365 days prior to the TRS date, such as personality disorders 24% (vs. 8% non‐TRS), episodic mood disorders 73% (vs. 48% non‐TRS), and self‐harm 3% (vs. 0.9% non‐TRS). The rates in the Stroup and clozapine groups were lower; for example, personality disorder was 18% and 19%, respectively. Patients with any of the listed psychiatric comorbidities were significantly more likely to be identified as having TRS than those without, as shown in Table 3. The registry TRS algorithm was significantly more likely to identify patients with TRS with any one of these comorbidities compared with those identified with the clozapine algorithm; for example, an episodic mood disorder diagnoses had an IRR of 4.51 (95% confidence interval, CI [4.26, 4.77]) for the registry TRS algorithm compared with an IRR of 3.22 (2.98, 3.46) for the clozapine algorithm. The Stroup algorithm was also sensitive at identifying these patients, significantly more so than the clozapine algorithm.
Table 3.
Rates of psychiatric comorbidity and hospital burden in patients with TRS in the United States
| Variable | Level | Stroup | Clozapine | Registry TRS | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Person years | TRS events | Crude rate | IRR | Person years | TRS events | Crude rate | IRR | Person years | TRS events | Crude rate | IRR | ||
| Gender of patient | Female | 463,492 | 10,816 | 0.0233 | Ref | 492,952 | 2,045 | 0.0041 | Ref | 497,644 | 3,253 | 0.0065 | Ref |
| Male | 500,928 | 11,266 | 0.0225 | 0.96 (0.94, 0.99) | 521,689 | 2,515 | 0.0048 | 1.16 (1.10, 1.23) | 537,707 | 3,335 | 0.0062 | 0.95 (0.90, 1.00) | |
| Age group | 13–17 | 24,732 | 1,343 | 0.0543 | 1.25 (1.17, 1.33) | 28,243 | 291 | 0.0103 | 1.09 (0.95, 1.24) | 28,578 | 287 | 0.0100 | 0.82 (0.72, 0.94) |
| 18–24 | 76,996 | 3,352 | 0.0435 | Ref | 85,004 | 807 | 0.0095 | Ref | 85,782 | 1,048 | 0.0122 | Ref | |
| 25–34 | 151,500 | 4,396 | 0.0290 | 0.67 (0.64, 0.70) | 161,312 | 1,056 | 0.0065 | 0.69 (0.63, 0.76) | 164,141 | 1,494 | 0.0091 | 0.75 (0.69, 0.81) | |
| 35–44 | 198,639 | 4,777 | 0.0240 | 0.55 (0.53, 0.58) | 209,664 | 907 | 0.0043 | 0.46 (0.41, 0.50) | 214,686 | 1,591 | 0.0074 | 0.61 (0.56, 0.66) | |
| 45–54 | 263,079 | 5,161 | 0.0196 | 0.45 (0.43, 0.47) | 273,685 | 952 | 0.0035 | 0.37 (0.33, 0.40) | 281,227 | 1,465 | 0.0052 | 0.43 (0.39, 0.46) | |
| 55–64 | 153,528 | 2,616 | 0.0170 | 0.39 (0.37, 0.41) | 159,879 | 471 | 0.0029 | 0.31 (0.28, 0.35) | 163,273 | 583 | 0.0036 | 0.29 (0.26, 0.32) | |
| 65–74 | 62,133 | 361 | 0.0058 | 0.13 (0.12, 0.15) | 62,931 | 66 | 0.0010 | 0.11 (0.09, 0.14) | 63,594 | 98 | 0.0015 | 0.13 (0.10, 0.16) | |
| 75+ | 33,810 | 76 | 0.0022 | 0.05 (0.04, 0.06) | 33,922 | 10 | 0.0003 | 0.03 (0.02, 0.06) | 34,066 | 22 | 0.0006 | 0.05 (0.03, 0.08) | |
| Race | Black | 417,059 | 7,408 | 0.0178 | 0.66 (0.64, 0.68) | 438,165 | 1,251 | 0.0029 | 0.49 (0.46, 0.53) | 440,070 | 2,433 | 0.0055 | 0.78 (0.74, 0.83) |
| Hispanic | 11,960 | 339 | 0.0283 | 1.06 (0.95, 1.18) | 12,395 | 89 | 0.0072 | 1.24 (1.00, 1.53) | 12,842 | 100 | 0.0078 | 1.10 (0.91, 1.35) | |
| Other | 67,827 | 1,813 | 0.0267 | 1.00 (0.95, 1.05) | 71,409 | 366 | 0.0051 | 0.88 (0.79, 0.99) | 72,446 | 458 | 0.0063 | 0.90 (0.81, 0.99) | |
| White | 467,573 | 12,522 | 0.0268 | Ref | 492,671 | 2,854 | 0.0058 | Ref | 509,992 | 3,597 | 0.0071 | Ref | |
| Psychiatric comorbidities in last 365 days | |||||||||||||
| Self‐harm | Yes | 6,255 | 602 | 0.0962 | 4.29 (3.96, 4.65) | 7,335 | 109 | 0.0149 | 3.36 (2.78, 4.07) | 7,049 | 224 | 0.0318 | 5.13 (4.49, 5.87) |
| Personality disorder | Yes | 439,188 | 15,122 | 0.0344 | 2.60 (2.53, 2.67) | 478,442 | 2,807 | 0.0059 | 1.79 (1.69, 1.90) | 478,869 | 4,820 | 0.0101 | 3.17 (3.00, 3.35) |
| Unipolar depression | Yes | 58,267 | 3,900 | 0.0669 | 3.34 (3.22, 3.45) | 67,676 | 852 | 0.0126 | 3.22 (2.98, 3.46) | 66,747 | 1,562 | 0.0234 | 4.51 (4.26, 4.77) |
| No | 906,152 | 18,182 | 0.0201 | Ref | 946,966 | 3,708 | 0.0039 | Ref | 968,604 | 5,026 | 0.0052 | Ref | |
| Burden | |||||||||||||
| Inpatient stay in last 365 days | 0 | 847,885 | 14,602 | 0.0172 | Ref | 884,932 | 2,921 | 0.0033 | Ref | 903,922 | 3,971 | 0.0044 | Ref |
| 1 | 90,439 | 4,904 | 0.0542 | 3.15 (3.05, 3.25) | 98,567 | 981 | 0.0100 | 3.02 (2.80, 3.24) | 100,049 | 1,470 | 0.0147 | 3.34 (3.15, 3.55) | |
| 2 | 17,546 | 1,515 | 0.0863 | 5.01 (4.76, 5.29) | 20,339 | 375 | 0.0184 | 5.59 (5.02, 6.22) | 20,697 | 568 | 0.0274 | 6.25 (5.72, 6.82) | |
| 3+ | 8,548 | 1,061 | 0.1241 | 7.21 (6.77, 7.67) | 10,803 | 283 | 0.0262 | 7.94 (7.02, 8.97) | 10,683 | 579 | 0.0542 | 12.34 (11.31, 13.46) | |
| Number of SZ hospitalizations days in the last year | 0 days | 851,917 | 14,776 | 0.0173 | Ref | 888,678 | 2,952 | 0.0033 | Ref | 907,655 | 4,042 | 0.0045 | Ref |
| 1–14 days | 103,424 | 6,288 | 0.0608 | 3.51 (3.40, 3.61) | 115,296 | 1,111 | 0.0096 | 2.90 (2.71, 3.11) | 116,558 | 1,979 | 0.0170 | 3.81 (3.61, 4.02) | |
| 15–29 days | 9,078 | 1,018 | 0.1121 | 6.47 (6.07, 6.89) | 10,668 | 497 | 0.0466 | 14.02 (12.75, 15.42) | 11,138 | 567 | 0.0509 | 11.43 (10.47, 12.48) | |
| Outpatient visit in last 365 days | 0 | 378,404 | 4,632 | 0.0122 | Ref | 398,609 | 790 | 0.0020 | Ref | 397,348 | 1,561 | 0.0039 | Ref |
| 1 | 135,924 | 4,157 | 0.0306 | 2.50 (2.40, 2.61) | 141,946 | 1,207 | 0.0085 | 4.29 (3.92, 4.69) | 142,033 | 968 | 0.0068 | 1.73 (1.60, 1.88) | |
| 2 | 62,702 | 1,648 | 0.0263 | 2.15 (2.03, 2.27) | 65,870 | 334 | 0.0051 | 2.56 (2.25, 2.91) | 66,327 | 523 | 0.0079 | 2.01 (1.82, 2.22) | |
| 3+ | 387,388 | 11,645 | 0.0301 | 2.46 (2.37, 2.54) | 408,215 | 2,229 | 0.0055 | 2.76 (2.54, 2.99) | 429,642 | 3,536 | 0.0082 | 2.09 (1.97, 2.22) | |
Note. IRR: incidence rate ratio; SZ: schizophrenia; TRS: treatment‐resistant schizophrenia.
In the Stroup group, it was not possible to assess hospital events in relation to TRS identification, as the algorithm required a hospital event to qualify, thus conditioning patients on hospitalization and potentially biasing quantification of these events. However, the data indicate that for patients in either the registry TRS or clozapine groups, patients with one or more hospital episodes were more likely to be identified as having TRS. For SZ patients with three or more hospital episodes, the registry TRS algorithm was 12 times more likely to identify them as having TRS compared with those with no episodes—IRR 12.34 (95% CI [11.31, 13.46]) and was more sensitive at identification than the clozapine algorithm—IRR 7.94 (95% CI [7.02, 8.97]). Similarly, increasing length of hospital stay increased the rate of TRS identified patients, particularly stays over 15 days, IRR 14.02 (12.75, 15.42) for clozapine and IRR 11.43 (10.47, 12.48) for registry TRS patients, respectively. Multiple outpatient episodes were also indicative of TRS designation, slightly more so when using the clozapine algorithm; for example, more than three visits had an IRR of 2.76 (2.54, 2.99) versus 2.09 (1.97, 2.22) for the registry algorithm.
4. DISCUSSION
This study confirmed that the registry TRS algorithm is a suitable alternative to using clozapine prescriptions to identify patients with TRS. Younger TRS patients in the Swedish dataset had consistently lower GAF scores than non‐TRS from SZ diagnosis indicating poor functionality, consistent with TRS (Lally et al., 2016). In U.S. data, the algorithm was particularly sensitive at detecting TRS patients with psychiatric comorbidities, a predictive factor for TRS (Schennach et al., 2012; Wimberley et al., 2016) and was also likely to quantify those with higher rates of hospital attendance as having TRS.
The Swedish results showed that patients with TRS had an SZ diagnosis approximately 10 years earlier than non‐TRS patients across all ages but nevertheless older than some other populations (33 years, SD 12.3), and loss of functionality was greater in those <45 years during observation. These findings support the earlier work from Lally et al. (2016) where English TRS patients were younger at the time of their first psychotic episode and SZ diagnosis (Demjaha et al., 2017; Lally et al., 2016), suggesting that early SZ onset may be a predictor of TRS. We observed that patients in Sweden, who were diagnosed with SZ at a high age, initially came into contact with psychiatric care on average 9.5 years prior to the first SZ diagnosis and received their first diagnosis of a psychotic disorder almost 2 years prior to the SZ diagnosis. Patients also started AP drug therapy on average 1.5 years prior to the first SZ diagnosis. The reason for this considerable delay is not clear, but initial diagnostic uncertainty early in disease and a general reluctance to label someone with an SZ diagnosis may be part of the explanation for the high age at diagnosis. In some cases, it may also be explained by subjects having migrated to Sweden with a preexisting diagnosis. We should be cautious in generalizing the finding of a younger age being associated with TRS to other settings until this is confirmed in other national samples.
The U.S. and Swedish data identified higher psychiatric comorbidity in TRS patients. Episodic mood disorders were particularly prevalent, and prior studies have identified that depression along with personality disorder are potential predictors of TRS and indicative of poor health outcomes in SZ patients generally (Schennach et al., 2012; Wimberley et al., 2016). The registry algorithm was particularly sensitive to hospital burden. TRS was likely to be identified when hospital burden was high, such as multiple hospital stays and longer admission periods. This finding supports a previous study, which identified that the patients with SZ with the poorest response to APs had the highest number of days in hospital (Schennach et al., 2012).
True treatment response is hard to define in a database study. “Non‐TRS” groups in this study include patients who did not have clinical evidence of TRS as defined by GAF scores and reduced burden but would have had varying levels of response to APs as prior studies have suggested a variety of trajectories of treatment outcome (Levine & Rabinowitz, 2010; Schennach et al., 2012). Because of this inability to validate TRS using a clinical endpoint to ascertain specificity, that is, true negatives, this study did not attempt to estimate TRS prevalence as a proportion of the SZ population. The numbers of TRS patients derived in this study are clearly lower than would be found in a clinical setting, where prevalence is often reported around 30% (Howes et al., 2012; NICE, 2014). We acknowledge this inability to clinically validate our findings as a limitation of the study, and future validation of the algorithm is required in electronic medical datasets with recourse to chart review.
Discontinuation of any AP due to toxicity, rather than lack of effect, was also hard to evaluate. However, the use of the clinically recommended AP dose (or 400 mg chlorpromazine equivalence) for 6–15 weeks, plus exclusion of patients who had a dose reduction >20% from average dose, was expected to have removed many of the patients who were switched due to adverse events, assuming those with toxicity issues would stop before 6 weeks. The U.S. data indicate that although the Stroup algorithm identifies more patients with TRS than the registry TRS algorithm, it is likely to also include those who stop due to toxicity or who are nonadherent, as there is no minimum or maximum time of exposure to APs included in the algorithm. Medication nonadherence is challenging to interpret in any claims database, as although prescriptions are filled, there is no evidence of actual drug use. The algorithm does attempt to deal with potential nonadherence by requiring a second prescription fill of AP, to at least imply persistence with therapy. But it is possible that a portion of those patients who are classified as TRS are nonresponsive due to a failure to take their medication, but this is not a problem that is unique to this targeting algorithm, it is an issue that is endemic to all pharmaceutical outcome research.
In summary, in this study we tested an algorithm that targets patients with TRS based upon clinical guidelines for TRS in patients with schizophrenia. The algorithm has demonstrated good face validity in that it identified patients with poorer functioning, greater psychiatric comorbidity, and greater hospital burden, indicative of TRS. We were not able to quantify prevalence of the TRS population due to lack of a clinical validation measure. Databases pose methodological limitations, but until clinical outcome data can adequately identify TRS in SZ populations or a specific diagnostic code for TRS is available to clinicians and researchers in diagnostic manuals, this algorithm is likely to add a vital tool for epidemiology and health outcomes research in this field.
AUTHOR CONTRIBUTIONS
All authors were involved in writing the publication and read and approved the final manuscript. Dr. Jönsson led the Swedish analyses, and Drs. Watson and Simonsen developed and conducted the U.S. analyses. Drs. Brain and Kymes contributed to the study development and publication.
FUNDING INFORMATION
This study was funded by the H. Lundbeck A/S.
DECLARATION OF INTEREST STATEMENT
Dr. Watson is an interim contractor to H. Lundbeck A/S and provides epidemiology support. She also takes fee from other pharmaceutical companies, but no conflicting financial interests or commercial interests are declared. Drs. Jönsson, Simonsen, and Brain are employees of H. Lundbeck A/S, and Dr. Kymes is an employee of Lundbeck, LLC.
ACKNOWLEDGEMENTS
Drs. R. Färdig and I. M. Wieselgren kindly contributed by extracting the data from the Swedish registries.
Jönsson L, Simonsen J, Brain C, Kymes S, Watson L. Identifying and characterizing treatment‐resistant schizophrenia in observational database studies. Int J Methods Psychiatr Res. 2019;28:e1778. 10.1002/mpr.1778
Footnotes
p < 0.05.
p < 0.01.
REFERENCES
- Brekke, J. S. , Prindle, C. , Bae, S. W. , & Long, J. D. (2001). Risks for individuals with schizophrenia who are living in the community. Psychiatric Services, 52(10), 1358–1366. 10.1176/appi.ps.52.10.1358 [DOI] [PubMed] [Google Scholar]
- Demjaha, A. , Lappin, J. M. , Stahl, D. , Patel, M. X. , MacCabe, J. H. , Howes, O. D. , … Murray, R. M. (2017). Antipsychotic treatment resistance in first‐episode psychosis: Prevalence, subtypes and predictors. Psychological Medicine, 47(11), 1981–1989. 10.1017/S0033291717000435 [DOI] [PubMed] [Google Scholar]
- Gardner, D. M. , Murphy, A. L. , O'Donnell, H. , Centorrino, F. , & Baldessarini, R. J. (2010). International consensus study of antipsychotic dosing. The American Journal of Psychiatry, 167(6), 686–693. 10.1176/appi.ajp.2009.09060802 [DOI] [PubMed] [Google Scholar]
- Hall, R. C. (1995). Global assessment of functioning. A modified scale. Psychosomatics, 36(3), 267–275. 10.1016/S0033-3182(95)71666-8 [DOI] [PubMed] [Google Scholar]
- Hambrecht, M. , Riecher‐Rossler, A. , Fatkenheuer, B. , Louza, M. R. , & Hafner, H. (1994). Higher morbidity risk for schizophrenia in males: Fact or fiction? Comprehensive Psychiatry, 35(1), 39–49. 10.1016/0010-440X(94)90168-6 [DOI] [PubMed] [Google Scholar]
- Hasan, A. , Falkai, P. , Wobrock, T. , Lieberman, J. , Glenthoj, B. , Gattaz, W. F. , … World Federation of Societies of Biological Psychiatry Task Force on Treatment Guidelines for Schizophrenia (2012). World Federation of Societies of Biological Psychiatry (WFSBP) guidelines for biological treatment of schizophrenia, part 1: Update 2012 on the acute treatment of schizophrenia and the management of treatment resistance. The World Journal of Biological Psychiatry, 13(5), 318–378. 10.3109/15622975.2012.696143 [DOI] [PubMed] [Google Scholar]
- Howes, O. D. , McCutcheon, R. , Agid, O. , de Bartolomeis, A. , van Beveren, N. J. , Birnbaum, M. L. , … Correll, C. U. (2017). Treatment‐resistant schizophrenia: Treatment Response and Resistance in Psychosis (TRRIP) working group consensus guidelines on diagnosis and terminology. The American Journal of Psychiatry, 174(3), 216–229. 10.1176/appi.ajp.2016.16050503 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Howes, O. D. , Vergunst, F. , Gee, S. , McGuire, P. , Kapur, S. , & Taylor, D. (2012). Adherence to treatment guidelines in clinical practice: Study of antipsychotic treatment prior to clozapine initiation. The British Journal of Psychiatry, 201(6), 481–485. 10.1192/bjp.bp.111.105833 [DOI] [PubMed] [Google Scholar]
- Iasevoli, F. , Giordano, S. , Balletta, R. , Latte, G. , Formato, M. V. , Prinzivalli, E. , … de Bartolomeis, A. (2016). Treatment resistant schizophrenia is associated with the worst community functioning among severely‐ill highly‐disabling psychiatric conditions and is the most relevant predictor of poorer achievements in functional milestones. Progress in Neuro‐Psychopharmacology & Biological Psychiatry, 65, 34–48. 10.1016/j.pnpbp.2015.08.010 [DOI] [PubMed] [Google Scholar]
- Jones, S. H. , & Castle, D. J. (2006). Management of treatment resistant schizophrenia. African Journal of Psychiatry, 9(1), 17–23. [Google Scholar]
- Jones, S. H. , Thornicroft, G. , Coffey, M. , & Dunn, G. (1995). A brief mental health outcome scale‐reliability and validity of the Global Assessment of Functioning (GAF). The British Journal of Psychiatry, 166(5), 654–659. 10.1192/bjp.166.5.654 [DOI] [PubMed] [Google Scholar]
- Kane, J. , Honigfeld, G. , Singer, J. , & Meltzer, H. (1988). Clozapine for the treatment‐resistant schizophrenic. A double‐blind comparison with chlorpromazine. Archives of General Psychiatry, 45(9), 789–796. 10.1001/archpsyc.1988.01800330013001 [DOI] [PubMed] [Google Scholar]
- Kelly, D. L. , Freudenreich, O. , Sayer, M. A. , & Love, R. C. (2018). Addressing barriers to clozapine underutilization: A national effort. Psychiatric Services, 69(2), 224–227. 10.1176/appi.ps.201700162 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kennedy, J. L. , Altar, C. A. , Taylor, D. L. , Degtiar, I. , & Hornberger, J. C. (2014). The social and economic burden of treatment‐resistant schizophrenia: A systematic literature review. International Clinical Psychopharmacology, 29(2), 63–76. 10.1097/YIC.0b013e32836508e6 [DOI] [PubMed] [Google Scholar]
- Lally, J. , Ajnakina, O. , Di Forti, M. , Trotta, A. , Demjaha, A. , Kolliakou, A. , … Shergil, S. S. (2016). Two distinct patterns of treatment resistance: Clinical predictors of treatment resistance in first‐episode schizophrenia spectrum psychoses. Psychological Medicine, 46(15), 3231–3240. 10.1017/S0033291716002014 [DOI] [PubMed] [Google Scholar]
- Lange, B. , Mueller, J. K. , Leweke, F. M. , & Bumb, J. M. (2017). How gender affects the pharmacotherapeutic approach to treating psychosis—A systematic review. Expert Opinion on Pharmacotherapy, 18(4), 351–362. 10.1080/14656566.2017.1288722 [DOI] [PubMed] [Google Scholar]
- Lehman, A. F. , Kreyenbuhl, J. , Buchanan, R. W. , Dickerson, F. B. , Dixon, L. B. , Goldberg, R. , … Steinwachs, D. M. (2004). The Schizophrenia Patient Outcomes Research Team (PORT): Updated treatment recommendations 2003. Schizophrenia Bulletin, 30(2), 193–217. 10.1093/oxfordjournals.schbul.a007071 [DOI] [PubMed] [Google Scholar]
- Leucht, S. , Samara, M. , Heres, S. , Patel, M. X. , Furukawa, T. , Cipriani, A. , … Davis, J. M. (2015). Dose equivalents for second‐generation antipsychotic drugs: The classical mean dose method. Schizophrenia Bulletin, 41(6), 1397–1402. 10.1093/schbul/sbv037 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Levine, S. Z. , & Rabinowitz, J. (2010). Trajectories and antecedents of treatment response over time in early‐episode psychosis. Schizophrenia Bulletin, 36(3), 624–632. 10.1093/schbul/sbn120 [DOI] [PMC free article] [PubMed] [Google Scholar]
- McIlwain, M. E. , Harrison, J. , Wheeler, A. J. , & Russell, B. R. (2011). Pharmacotherapy for treatment‐resistant schizophrenia. Neuropsychiatric Disease and Treatment, 7, 135–149. 10.2147/NDT.S12769 [DOI] [PMC free article] [PubMed] [Google Scholar]
- National Institute for Health and Care Excellence . (2014). Psychosis and schizophrenia in adults: Treatment and management: Clinical guideline no. 178. Retrieved from London: [PubMed]
- Nielsen, J. , Young, C. , Ifteni, P. , Kishimoto, T. , Xiang, Y. T. , Schulte, P. F. , … Taylor, D. (2016). Worldwide differences in regulations of clozapine use. CNS Drugs, 30(2), 149–161. 10.1007/s40263-016-0311-1 [DOI] [PubMed] [Google Scholar]
- Novartis Pharmaceutical Corporation (2017). Clozaril prescribing information 2017. Retrieved from
- Schennach, R. , Meyer, S. , Seemuller, F. , Jager, M. , Schmauss, M. , Laux, G. , … Klosterkötter, J. (2012). Response trajectories in “real‐world” naturalistically treated schizophrenia patients. Schizophrenia Research, 139(1–3), 218–224. 10.1016/j.schres.2012.05.004 [DOI] [PubMed] [Google Scholar]
- Siskind, D. , Siskind, V. , & Kisely, S. (2017). Clozapine response rates among people with treatment‐resistant schizophrenia: Data from a systematic review and meta‐analysis. Canadian Journal of Psychiatry, 62(11), 772–777. 10.1177/0706743717718167 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Strassnig, M. T. , & Harvey, P. D. (2014). Treatment resistance and other complicating factors in the management of schizophrenia. CNS Spectrums, 19(Suppl 1), 16–23. quiz 13–15, 24. 10.1017/S1092852914000571 [DOI] [PubMed] [Google Scholar]
- Stroup, T. S. (2014). What is the role of long‐acting injectable antipsychotics in the treatment of schizophrenia? The Journal of Clinical Psychiatry, 75(11), 1261–1262. 10.4088/JCP.14com09518 [DOI] [PubMed] [Google Scholar]
- Stroup, T. S. , Gerhard, T. , Crystal, S. , Huang, C. , & Olfson, M. (2016). Comparative effectiveness of clozapine and standard antipsychotic treatment in adults with schizophrenia. The American Journal of Psychiatry, 173(2), 166–173. 10.1176/appi.ajp.2015.15030332 [DOI] [PubMed] [Google Scholar]
- Wimberley, T. , Stovring, H. , Sorensen, H. J. , Horsdal, H. T. , MacCabe, J. H. , & Gasse, C. (2016). Predictors of treatment resistance in patients with schizophrenia: A population‐based cohort study. Lancet Psychiatry, 3(4), 358–366. 10.1016/S2215-0366(15)00575-1 [DOI] [PubMed] [Google Scholar]