Time Course of Clozapine-associated Agranulocytosis Over 19 Years

Background. Clozapine is the most effective antipsychotic for treatment-resistant schizophrenia, yet its use is limited by the risk of agranulocytosis, a rare but life-threatening adverse effect that led to mandatory hematological monitoring. In the original U.S. surveillance cohort (Alvir et al., 1993), cumulative incidence was 0.80% at one year and 0.91% at 18 months, with most cases occurring within the first three months. More recent meta-analytic estimates are lower (approximately 0.4%). The concentration of risk early in treatment underlies monitoring protocols that reduce testing frequency after an uneventful first year. However, whether risk approaches zero thereafter remains uncertain: in a 25-year Finnish national cohort, more than one in ten reported cases occurred after the second year of treatment, and the clinical significance of age and of late-onset events is unresolved. In Israel, clozapine prescriptions are limited to a monthly supply and require a monthly blood count for renewal. This allows consistent long-term follow-up through the Sheba Medical Center data warehouse.

Objectives. The primary objective is to characterize the full time course of agranulocytosis among clozapine-treated patients over up to approximately 19 years, and to estimate cumulative incidence and period-specific incidence rates, distinguishing the early window (first 18 weeks) from subsequent follow-up. The secondary objectives are:

To examine whether the incidence rate after 18 weeks is stable or changes with treatment duration.

To assess age, sex and psychiatric diagnosis as predictors within 48 months and across the full follow-up.

To distinguish, through structured case-level adjudication, clozapine-attributable events from neutropenia related to malignancy or cytotoxic chemotherapy, and from events reflecting an erroneous index date.

To test robustness to assumptions about censoring, left truncation and inclusion of patients with malignancy.

Design and data source. This is a retrospective, non-interventional cohort study based on electronic health records from the Sheba Medical Center data warehouse covering February 2007 to August 2026. The following data will be extracted:

ambulatory prescriptions and chronic medication lists (ATC N05AH02 and concomitant medications) absolute neutrophil counts (ANC) ICD-9 diagnosis codes psychiatric clinic visits demographics

Exposure and follow-up. The index date is the first documented clozapine exposure. Exposure time is defined by consecutive monthly renewal months with a documented psychiatric clinic visit. Follow-up ends at the earliest of the following: agranulocytosis, the last consecutive renewal month, or the end of the analysis window.

Outcome. The primary outcome is agranulocytosis, defined as at least two ANC measurements below 0.5×10⁹/L within one week. The event date is the first qualifying measurement. The primary analysis window extends to 48 months, and an extended analysis covers all available follow-up.

Case-level adjudication. All laboratory-defined cases will undergo structured review of the following:

timing relative to treatment start neutrophil trajectory malignancy and oncologic or hematologic treatment codes

Events consistent with cytotoxic chemotherapy, and events occurring within one day of the first community prescription (likely inpatient initiation), will be excluded from the primary (adjudicated) cohort but retained in the unadjudicated cohort for comparison.

Statistical analysis. The analysis comprises the following:

Crude risk with Clopper-Pearson confidence intervals. Incidence rates per 1,000 person-years with Poisson confidence intervals across the intervals 0-18 weeks, 18 weeks-6 months, 6-12, 12-24, 24-60, 60-120 and >120 months.

Poisson regression and Weibull survival modeling to assess rate stability after 18 weeks.

Kaplan-Meier cumulative incidence with log-log confidence intervals. Firth penalized logistic regression for predictors, with an unpenalized Cox model as a check.

Sensitivity analyses. Four sensitivity analyses are planned:

Excluding all patients with any malignancy code. Restricting to first exposure in 2009 or later (to address left truncation). An actuarial life table using last clinical contact as the end of observation. Comparing the adjudicated and unadjudicated cohorts.

Ethics. The study will be submitted to the Sheba Medical Center Helsinki Committee (protocol 3658-26-S) with a request for a waiver of informed consent. Data will be extracted in coded form without direct identifiers, stored on secured hospital servers, and reported only in aggregate. Cells with fewer than five patients will not be reported in an identifiable manner.