Mortality and Self-Harm in Association With Clozapine in Treatment-Resistant Schizophrenia

We extracted information on medication from the Danish National Prescription Registry, where all outpatient drug prescriptions have been registered since 1995 (18). We obtained information on admission dates and diagnoses (ICD versions 8 and 10) from the Danish National Patient Registry (19) and the Danish Psychiatric Central Research Register (20). We obtained information on sex, date of birth, vital status, and nationality and parents' personal identification numbers from the Danish Civil Registration System (21). Information on causes of death was obtained from the Causes of Death Register; information was available until Dec. 31, 2011 (22). The unique personal identification number was used to link individual data across the national registration systems, including registers holding sociodemographic information (21).

We conducted a population-based cohort study. The cohort comprised all individuals born in Denmark after Jan. 1, 1955, with a first diagnosis of schizophrenia (ICD 8: 295.x9, excluding 295.79; ICD-10: F20) at age 18 or older and after Jan. 1, 1996, and fulfilling criteria for treatment resistance before June 1, 2013. To define the cohort and start of follow-up (baseline), we used a register-based definition of treatment-resistant schizophrenia based on either of the following two criteria: 1) clozapine prescription filled at a pharmacy or 2) psychiatric hospital admission within 18 months during continued treatment with antipsychotics after at least two periods of monotherapy with different antipsychotics, each lasting at least 6 weeks. The definition has been applied and described more fully elsewhere (23).

We studied all-cause and cause-specific mortality as well as the first recorded episode of self-harm after the patient met one of the criteria for treatment resistance. All-cause mortality was defined by using the recorded date of death retrieved from the Danish Civil Registration System, where vital status is continuously updated. We assessed causes of death from the Causes of Death Register: suicide, death from diseases and medical conditions, and death from other external causes, as classified in a previous study (24). Self-harm was defined as the first registered episode of self-harm after the criteria for treatment resistance were met. Self-harm events included suicide attempts as well as self-harm behaviors such as cutting and poisoning (excluding use of food, alcohol, and mild analgesics) (25–28).

The primary exposure was defined as time-varying treatment, classifying individual follow-up time into periods of clozapine treatment and no clozapine treatment. This was defined from prescription data as described in Table S1 in the data supplement accompanying the online version of this article. “No clozapine treatment” was further classified into nonclozapine antipsychotic treatment and no antipsychotic treatment so that the former could serve as an active comparator group.

Secondary exposure measures were also defined. First, to account for periods of concomitant antipsychotic treatment, treatment status was classified according to the following subcategories: clozapine monotherapy (reference), clozapine with other antipsychotics, nonclozapine antipsychotic polypharmacy, nonclozapine antipsychotic monotherapy, and no antipsychotic treatment. Next, to study the timing of all-cause mortality in relation to clozapine, the time-varying treatment was classified as periods of past, no, and current clozapine treatment.

We adjusted for sex and previous episodes of self-harm as well as the following time-dependent factors: age, calendar year, comorbid substance abuse, comorbid somatic disorders (Charlson index score >0) (29), comorbid psychiatric diagnoses (other schizophrenia spectrum disorder, singular or recurrent depression, personality disorder), living in the capital area, and cumulative clozapine treatment (0, 0–1, 1–3, ≥3 years).

We performed crude and adjusted Cox proportional hazards regression and analyzed time to death as well as time to the first recorded episode of self-harm in separate models, where individuals were followed from the date of meeting criterion 1 or 2 for treatment-resistant schizophrenia, whichever came first. Individuals were censored at emigration from Denmark or end of follow-up (June 1, 2013). Clozapine treatment (the recommended treatment in treatment-resistant schizophrenia) was used as the reference, allowing for direct presentation of hazard ratios for several comparisons, including different antipsychotic treatment strategies. This implies, for instance, that a hazard ratio above 1 favors clozapine, as it has a lower event rate. Estimates for the opposite comparison can easily be obtained by inversion of both the hazard ratio and confidence limits. For all-cause mortality, we additionally studied the timing of death by comparing rates after clozapine discontinuation (designated as past clozapine treatment) and rates during nonclozapine treatment with rates during current clozapine treatment. For this analysis, we reset the time period at the beginning or end of a clozapine treatment period. Because individuals could contribute to several treatment periods, we used robust standard errors to account for intraindividual correlation. Hazard ratios were presented for models splitting follow-up into the following time intervals: 0–1, 1–3 and ≥3 years to assess the timing of death after clozapine discontinuation. For cause-specific death, analyses were conducted for crude and partly adjusted models and with further adjustment for psychiatric hospitalization in the previous year. Individuals were censored at death from other causes, emigration from Denmark, or last available information on cause of death from the Causes of Death Register (Dec. 31, 2011).

The proportional hazards assumption for the Cox regression models was evaluated from diagnostic plots for baseline variables. All estimates are accompanied by 95% confidence intervals (CIs). All analyses were conducted in Stata version 13 (StataCorp, College Station, Tex.).

We identified 2,370 individuals meeting the criteria for treatment-resistant schizophrenia. Characteristics are shown in Table 1; 45.8% were women, and the median age at the point of meeting criteria for treatment resistance was 30.1 years (interquartile range [IQR]: 24.8–37.3). In total, 158 (6.7%) died, and 602 (25.4%) had at least one episode of self-harm registered during follow-up (maximum of 17 years, median: 6.8 years, IQR: 3.2–10.6). The overall rates per 100 person-years were 0.9 (95% CI: 0.8–1.1) for all-cause death and 4.6 (95% CI: 4.2–5.0) for self-harm. Rates of death, as well as rates of self-harm, differed across most baseline characteristics (Table 1). During follow-up, 1,372 individuals (58%) with treatment-resistant schizophrenia initiated clozapine treatment. Among these, the median number of clozapine treatment periods was 3, and the median duration of a single clozapine treatment period was 10 months (IQR: 5–19 months). Clozapine users were younger and more likely to have been admitted to a psychiatric hospital within the previous year, but they were less likely to have had a diagnosis of substance abuse and to have lived in the capital area (Table 1).

When follow-up was divided by treatment status, the distribution of risk time was as follows: clozapine monotherapy, 15%; clozapine with other antipsychotics, 17%; nonclozapine antipsychotic monotherapy, 26%; other antipsychotic polypharmacy, 16%; and no antipsychotic treatment throughout follow-up, 26%.

In a Cox regression model with time-varying clozapine treatment, we found that the absence of clozapine treatment was associated with an elevated rate of all-cause mortality (hazard ratio: 1.88, 95% CI: 1.16–3.05), compared with clozapine treatment in adjusted models. Estimates were substantially higher for no antipsychotic treatment (hazard ratio: 2.50, 95% CI: 1.50–4.17) and somewhat higher for nonclozapine antipsychotic treatment (hazard ratio: 1.45, 95% CI: 0.86–2.45) than for clozapine treatment (Table 2, model B). When we compared different treatment strategies with clozapine monotherapy, the absence of antipsychotic treatment was again associated with the highest rate of death compared with clozapine treatment (hazard ratio: 2.66, 95% CI: 1.36–5.18), whereas no significant differences were seen between clozapine monotherapy and any other antipsychotic treatment strategy. The hazard ratio estimate comparing nonclozapine antipsychotic monotherapy with clozapine monotherapy was high, although nonsignificant (hazard ratio: 1.75, 95% CI: 0.87–3.50) (Table 2, model B).

Rates of all-cause mortality were highest after clozapine discontinuation, particularly within the first year after clozapine discontinuation when compared with rates during clozapine treatment (Figure 1). Compared with current clozapine treatment, the 1-year mortality hazard ratios adjusted for sex, age, calendar year, psychiatric hospitalization, and comorbid somatic diseases were 1.41 (95% CI: 0.63–3.13) in periods of no past or current clozapine treatment and 2.65 (95% CI: 1.47–4.78) in periods after clozapine discontinuation.

The analyses with cause-specific mortality included 2,141 individuals, of whom 125 (5.8%) died during follow-up. The estimated adjusted hazard ratios for nonclozapine antipsychotic treatment or no antipsychotic-based treatment at all were, with clozapine treatment used as the reference, 1.74 (95% CI: 0.59–5.12) for suicide, 1.41 (95% CI: 0.80–2.49) for deaths from other diseases or medical conditions, and 0.77 (95% CI: 0.31–1.89) for deaths from other external causes (Table 3, model C).

Nonclozapine antipsychotic treatment was associated with an elevated rate of self-harm; the hazard ratio was 1.36 (95% CI: 1.04–1.78), compared with clozapine in adjusted models, whereas no association was found when no antipsychotic treatment was compared with clozapine (hazard ratio: 1.15, 95% CI: 0.86–1.53) (Table 4, model B). The estimated rates of self-harm were lowest for clozapine (monotherapy or polypharmacy) and highest for nonclozapine antipsychotic polypharmacy, but no significant differences were found across treatment strategies in any of the models (Table 4).

When periods of psychiatric hospitalization lasting longer than 1 month were excluded, effect sizes were slightly decreased in most comparisons, but the conclusion was unaltered in all models (Table 2, Table 4). The analyses censoring at first change in treatment status resulted in a substantially increased hazard ratio of 3.58 (95% CI: 1.14–11.27) for all-cause mortality (crude analysis due to few events) and an adjusted hazard ratio of 3.08 (95% CI: 1.25–7.57) for self-harm. Analyses of long-acting injectable antipsychotics compared with clozapine resulted in an adjusted hazard ratio of 1.28 (95% CI: 0.72–2.26) for all-cause mortality and an adjusted hazard ratio of 1.24 (95% CI: 0.99–1.79) for self-harm (Table S2 in online data supplement). Results of the initial-treatment approach are shown in the online supplementary material (Tables S3 and S4).

A major strength of the present study is the population-based, longitudinal study design linking several registers to gather information obtained at baseline as well as multiple times over a long follow-up period, up to 17 years. Information on all medication dispensed at all pharmacies in Denmark since 1995 was available, and the fact that the individual received the medication from the pharmacy largely indicates adherence. Another strength of the study is the restriction of the cohort to individuals meeting criteria for treatment resistance, which we consider the most appropriate approach for studying outcomes in relation to clozapine treatment, because all members of the cohort are considered to have an indication for clozapine. The study population used by Stroup and colleagues was also restricted to individuals with treatment-resistant schizophrenia (15), but unlike them, we additionally used a time-varying treatment design, which enabled us to study the temporal and acute treatment effects over long-term follow-up, adjusting for several time-dependent factors associated with treatment and outcome. As a consequence of our chosen design, we assigned periods of no antipsychotic prescriptions to a separate exposure category.

Observational studies using registry data have some limitations. One limitation is the lack of information on antipsychotic medication status during hospitalization. However, we repeated analyses excluding periods of psychiatric hospitalization exceeding 1 month, resulting in similar estimates. Also, for a minor proportion of individuals, antipsychotic medication has been dispensed for up to 2 years free of charge through hospital pharmacies since 2008, resulting in no antipsychotic prescriptions being found for these individuals during this period. This might have biased the estimates, but with follow-up restricted to 2007, analyses resulted in similar estimated effect sizes.

The current study compared clozapine treatment with different overall treatment strategies but not with specific antipsychotic drugs, as was the case in the FIN-11 study (10). However, the current study is to our knowledge the first study to compare clozapine with other antipsychotic monotherapy and long-acting injectable antipsychotics in treatment-resistant schizophrenia.

The outcome in the present study termed “self-harm” included both suicide attempts (the majority) and self-harm without the intent of suicide. As we could not distinguish between the two and since underreporting of self-harm and suicide attempts from the Danish registers is a limitation (28), it is possible that clozapine does not have similar effects on the risk of self-harm (without intent of suicide) and the risk of suicide attempt.

Furthermore, the results could be biased due to uncontrolled confounding by such elements as symptom type and severity, nonantipsychotic comedication and psychotherapy, and the inability of the register-based design to distinguish between discontinuation of antipsychotic medication due to medication nonresponse, intolerance, and nonadherence. Similarly, the register-based definition of treatment resistance is not a perfect proxy because the registers available did not include information on treatment nonresponse, and our cohort might thus include some individuals who were intolerant rather than nonresponsive to treatment. Furthermore, we could not take into account the regular contact with the health care system due to clozapine blood monitoring, which could serve as a potential intermediate factor associated with decreased severity in clozapine initiators (12). However, clozapine remained the treatment associated with the lowest rates of death and self-harm in sensitivity analyses compared with long-acting injectable antipsychotics, which similarly require regular contacts with the health care system. No significant differences were observed between long-acting injectable antipsychotics and other nonclozapine antipsychotics, corroborating the findings of a recent meta-analysis of randomized controlled trials (33). Still, clozapine blood monitoring might, at least in part, explain the lower rate of deaths in the first year after clozapine initiation or reinitiation and the higher rate after clozapine discontinuation. Thus, this study explores the overall real-world effect of different antipsychotic treatments rather than exclusively the pharmacological effect of the drugs.

Finally, the design of time-varying treatment and confounders might introduce collider-stratification bias, i.e., bias due to conditioning on factors affected by both the prior status of the covariate as well as by the prior treatment status. However, alternative methods such as time-varying propensity scores or inverse probability weighting of marginal structural models may not improve results markedly (34).

The results of the present study indicate that clozapine use is associated with a decreased mortality rate, in line with previous research (10–12). This was, however, only significant when clozapine treatment was compared with periods of no antipsychotic treatment, a situation probably largely explained by an excess mortality rate observed after clozapine discontinuation. Furthermore, the results of the present study suggest a protective effect of clozapine in the prevention of self-harm when compared with other antipsychotics, but no effect was found when clozapine treatment was compared with no use of antipsychotics. It remains unclear whether the protective effect of clozapine on self-harm in treatment-resistant schizophrenia could be partly explained by a potentially harmful effect of alternative treatment strategies with other antipsychotics or confounding by indication. Moreover, the extent to which the observed excess mortality rate after clozapine discontinuation is caused by side effects from recent clozapine exposure, unobserved factors, or clozapine discontinuation remains to be investigated. This study suggests that clozapine discontinuation needs more attention with thorough evaluation, care, and monitoring of the patient.