Rare but elevated incidence of hematological malignancy after clozapine use in schizophrenia: a population cohort study

Data source

We adopted a retrospective cohort study design for this study. The Hospital Authority of Hong Kong, which is the sole provider of public inpatient services and a major provider of public outpatient services, provided strictly anonymized and de-identified electronic health records for data analysis. As all legal residents of Hong Kong are eligible for receiving services from the public sector, the database essentially covers the entire population and all territories of Hong Kong. The electronic health records were extracted from the Clinical Management System in which clinicians input routine patient attendance, diagnosis and medication records daily, linked by a unique person identity number. Diagnoses are coded based on the International Classification of Diseases, Ninth Revision, Clinical Modifications (ICD-9-CM), while medications are coded with British National Formulary codes as well as generic names, with dosage and duration information also available. Timestamps are available for every record entry, facilitating the delineation of the temporality of records and events. Many pharmacoepidemiologic research studies have already been published using this database,^13-16 with the accuracy of diagnostic codes well established.¹⁷

Participants

We included anonymized patients aged 18 years or older with a diagnosis of schizophrenia (ICD-9-CM code 295) and a record of using clozapine or olanzapine for a duration of 90 days or more in January 2001 -August 2022. The first prescription of clozapine or olanzapine date was designated as the index date. Patients who i.) did not use at least two other antipsychotics before the index date, ii.) those who ever used both clozapine and olanzapine in their prescription history, and iii.) those who had a prior record of any cancers(including HM) before the index date were excluded. Records dating back to the year 1999 were used to execute these exclusion criteria. Patients were followed from the index date (i.e., 90 days or more clozapine or olanzapine use) until i.) the diagnosis of HM, ii.) death, iii.) five years after the cessation of olanzapine or clozapine use, or iv.) end of data availability (i.e., August 31^st 2022), whichever came earliest.

Ethics approval for this study was obtained from the Institutional Review Board of the University of Hong Kong/ Hospital Authority Hong Kong West Cluster (HKU/HA HKW IRB, reference number: UW 20-113). Informed consent was waived as a requirement for ethics approval because the data were all anonymized.

Outcomes

Any HM diagnosis (ICD-9-CM codes 200-209, 238.4, 238.5, 238.6, 238.7, detail in eTable 1) as a composite measure was adopted as the outcome of this study.

Exposure and comparator

Clozapine and olanzapine were identified by their generic names in the electronic health records. We did not consider a 7-day, or a shorter, gap between repeated prescriptions of antipsychotics as a discontinuation given a half-life of olanzapine ranging from 21 to 54 hours.¹⁸ Clozapine was the primary exposure in this study while olanzapine was the comparator.

Statistical analysis

Poisson regression was used to estimate the incidence rate ratio (IRR) with 95% confidence intervals (CI) of HM between clozapine and olanzapine users, with an offset term to account for varying follow-up times for each patient. The inverse probability of treatment weighting (IPTW), using a propensity score, was applied to balance the characteristics between the clozapine and olanzapine users. The score was estimated by the logistics regression, considering covariates including age, sex, previous use of categories of antipsychotics according to their chemical structure (phenothiazines, thioxanthene derivatives, diphenylbutylpiperidine derivatives, butyrophenone derivatives, benzamides, indole derivatives, diazepines, oxazepines, thiazepines and oxepines, and other antipsychotics), prior mental disorder diagnoses (depression, bipolar disorder, and dementia), number of antipsychotic used in the past year, and immunological disease history (autoimmune disease and acquired immune deficiencies) which have been proven to associate with the HM also used as covariates^19,20. eTable 1 shows the ICD-9-CM codes and generic drug names used to ascertain the covariates. The standardized mean difference (SMD) was used to identify potential imbalances between clozapine and olanzapine users. Covariates with an SMD greater than 0.1 after IPTW were further adjusted in the regression model. In addition to IRRs, weighted absolute rate differences were also estimated between the two groups.

Subgroup analyses by age group and sex were conducted to provide specific estimates for different demographic strata. Additionally, several sensitivity analyses were conducted to test the robustness of our results: i) the cohort was further restricted to patients using clozapine or olanzapine for 180 days (instead of 90); ii) patients were censored for three years (instead of five) after their cessation of clozapine or olanzapine use; iii) a multivariable Poisson regression to employed to replicate the analysis; iv) a Cox regression with the IPTW approach was used to estimate the hazard ratio; and v) asthma (ICD-9-CM codes 493) and other cancers (ICD-9-CM codes 140-239, excluding HM) were used as negative control outcomes to check for any unmeasurable selection bias, as it shared the same potential source of bias with HM but was not causally related to antipsychotic exposure.

YH and LG were independent analysts of this study to minimize errors in coding and programming.

Cohort characteristics

Table 1 shows the baseline characteristics of the cohort before and after applying IPTW. Before weighting, the mean age of the clozapine group (41.00 ± 15.02) was approximately five years younger than that of the olanzapine group. The olanzapine group had a higher proportion of men (54.0%) than women, while clozapine the group had a smaller proportion of men (46.6%). On average, the follow-up time of the clozapine users were 4540.90 days compared to 2664.43 days follow-up time in olanzapine users, and more clozapine users prescribed more than three kinds of other antipsychotics in the past year (59.5% versus 36.9%). Notable differences were observed in the types of previous antipsychotics use. Specifically, phenothiazines (82.6% versus 64.6%) and thioxanthene derivatives (52.8% versus 37.4%) were more often used previously among clozapine users than olanzapine users. However, more olanzapine users had used diazepines, oxazepines, thiazepines, and oxepines (39.5% versus 28.4%), and ‘other antipsychotics’ (71.9% versus 57.3%). Additionally, a higher prevalence of psychiatric comorbidities and immunological diseases was observed among olanzapine users. After weighting, the SMD of characteristics were all less than 0.1 except prior benzamide use, which was adjusted in the subsequent weighted regression analysis.

Main analysis

Figure 2. shows the weighted cumulative incidence curves of HM over the follow-up period for clozapine and olanzapine users. The incidence of HM among clozapine users was consistently higher than that among olanzapine users for nearly the entire follow-up period.

• Download figure
• Open in new tab

Figure 2.

Cumulative incidence curves over the observation period by clozapine or olanzapine use status. Yellow represents clozapine and blue represents olanzapine.

Among all 9,965 individuals exposed to olanzapine or clozapine, a total of 39 individuals were diagnosed with HM. Out of the 39 cases, nine individuals were clozapine users (n = 834), while the remaining 30 individuals were olanzapine users (n = 9,131). Six clozapine users and 23 in olanzapine users developed HM after discontinuing the medication. For the six clozapine users, the average number of days between their discontinuation and HM was 256.67 days. For the 23 olanzapine users, the number was 713.86 days.

Results from the Poisson regression models show that compared to olanzapine users, clozapine users had a significantly higher IRR of 2.22 (95% CI 1.52, 3.34) for HM (Table 2). The absolute rate difference of HM between the two groups was 57.40 (95% CI 33.24, 81.55) per 100,000 person-years.

Subgroup analyses showed a significantly elevated risk of HM among clozapine users for all sex and age groups, except for those aged 65 or older despite a comparable IRR.

Sensitivity analysis

Sensitivity analyses yielded similar results (eTable 2). The IRR for asthma as a negative control outcome was not significant and close to one (IRR 0.90, 95% CI 0.66,1.21). Likewise, the IRR for other cancers was estimated at 0.77 (95% CI 0.68, 0.86).

Relationship with the literature

To the best of our knowledge, this work is the world’s first analytic cohort study on this association to estimate absolute rate differences, with only two prior notable case-control studies conducted in Finland⁷ and the United States.⁸ Although the Finnish study also provided descriptive data on the incidence of HM using a cohort design, no hypothesis testing was conducted in the cohort study. Indeed, our descriptive absolute risk estimates were highly comparable with the Finnish study’s descriptive cohort.⁷ The rarity of HM and the relatively low prevalence of clozapine use compared to some other widely used antipsychotics have both greatly limited previous research on this association, despite compelling evidence on the association between clozapine use and a range of hematological abnormalities.^21-23 In fact, apart from this and the two case-control studies, existing evidence on an association with HM arose only from passive reporting systems using disproportionality analysis which may be subject to more biases than large electronic health record epidemiologic studies due to known limitations in adverse event reporting platforms and databases.^5,24 The previous case-control studies, despite the strengths of the national registry and comprehensive veteran health database, used retrospectively ascertained long-term clozapine use as the main exposure and ‘little to no such use’ as the comparator. This approach is not without limitations, as clozapine use is most typically preceded by a range of other antipsychotic use in the individuals because it is widely regarded as the most efficacious drug available.²⁵ Our current study address this challenge by confining our cohort to those who used two other antipsychotics before using clozapine (or olanzapine) and employing olanzapine as an active comparator.²⁶ This ensures a more proper adjustment of previous antipsychotic use before clozapine or olanzapine initiation. With a cohort design, we also estimated the absolute rate difference which is highly relevant in clinical decision making. The current study represented evidence that is one level higher up the clinical evidence pyramid compared with the existing knowledge, i.e., from case-control to cohort evidence.

Interpretation of observed association

The strong association observed in this and the previous two case-control studies⁷ as well as its good coherence with previous evidence on the hematological abnormalities²⁷ associated with clozapine constitute a strong case that this association is likely causal. We believe the effects of such abnormalities being induced by the prolonged use of clozapine may accumulate over the duration of use and potentially increase the risk of malignancy through a range of potential mechanisms.²⁸ For instance, recent evidence showed that the elevated risk of agranulocytosis following clozapine use may persist well over several years, and this mechanism could potentially impact the increased risk of other blood-related disorders among clozapine users, including a higher risk of HM.²⁹ We did not have a sufficient number of cases to separately analyze the type of HM with a more significantly elevated risk; inference about the specific underlying mechanism is thus limited. Further research should investigate specific HM types.

Potential clinical implications

Despite a small rate difference, a potential strong association between clozapine use and HM, if substantiated by further evidence, would provide additional information to the safety profile of the drug compared to olanzapine. This potential slight recalibration of the risk-benefit balance may partially impact clinical decisions about long-term clozapine use. First, considering the markedly elevated risk of HM, despite its rare incidence, psychiatrists may be more inclined to try other antipsychotics before initiating clozapine, which is currently the most efficacious antipsychotic drug. Second, for individuals with significant hematological abnormalities detected early in the course of clozapine use, alternative drugs may be beneficial, as prolonged use of clozapine may be associated with a higher chance of HM. Third, patients with a family history of HM may benefit from reconsideration of clozapine prescription, as they have a higher baseline risk.³⁰ These potential considerations need to be supported by additional evidence, such as studies specific to individuals with a family history of HM and the relationship between detected hematological abnormalities and subsequent HM risks.

Strengths and limitations

One of the key strengths of this study is the cohort design with a range of important covariates properly adjusted, e.g., prior antipsychotic use, providing highly useful information on the absolute rate of HM and the rate difference. Additionally, the territory-wide database with comprehensive medical records based on the same coding system and practices also confers great strengths, such as good generalizability and representativeness of the findings.

There are, nevertheless, limitations that warrant caution. First, this evidence is observational without randomization, which may entail potential selection or indication bias. Nevertheless, blood tests prior to the prescription of clozapine should have already excluded people with a higher risk of HM; such indication bias, if any, might lead to an underestimation of the observed association. Moreover, it is difficult to conduct randomized prospective studies on such a rare outcome, so future research should likely still be focused on observational studies with multiple sites or even countries. Second, although olanzapine, particularly in high doses, can be useful for some cases, it is not a suggested replacement option for clozapine in the treatment-resistance schizophrenia due to the lack of a highly comparable indication.³¹ Third, there are unobserved covariates such as socioeconomic status and lifestyle factors that were not included in this study. Fourth, there may be a detection bias of HM for clozapine, as it is known to be associated with hematological abnormalities which are closely monitored.³² However, since we allowed the observation to last until five years after discontinuation to account for symptomatic presentation, the detection bias should not be significant. Also, the association is strong and unlikely to be purely arising from detection bias, i.e., earlier detection. In fact, our findings showed that a smaller proportion of HM cases was recorded during the use of clozapine than that during the use of olanzapine. Fifth, the dosage of treatment was not standardized during the course of treatment or across individuals and is subject to the psychiatrists’ clinical judgement for each individual. Sixth, the follow-up time for the olanzapine group was shorter than that of the clozapine group, but it should likely be long enough to observe and capture the study outcome, with a mean follow-up time of more than seven years. Seventh, among clozapine users, many had used olanzapine before switching to clozapine in clinical practice but we excluded approximately 5,000 patients who had a history of using olanzapine either before or after starting clozapine treatment. This exclusion has limited our sample size although we arrived at a more selective sample for a fair comparison. Finally, the study population in Hong Kong is predominantly ethnic Chinese, and the generalizability of the findings to other populations need to be tested using multinational data.