Principal findings

This study provides the first large-scale quantitative data on clozapine initiation patterns using operationalised TRS criteria in the UAE. The 7.4% initiation rate is consistent with international reports of clozapine underutilisation, with around 10% in Danish national registers. ^{Reference Nielsen, Røge, Schjerning, Sørensen and Taylor20} The median delay of 13.2 months exceeds guideline recommendations, but is substantially shorter than the 4-year delays historically reported in the UK ^{Reference Howes, Vergunst, Gee, McGuire, Kapur and Taylor9} and comparable to 12-month delays in Australian early intervention cohorts. ^{Reference Thien, Bowtell, Eaton, Bardell-Williams, Downey and Ratheesh10} Our concordance rate of 26.3% mirrors the 28% reported in New Zealand, ^{Reference Wheeler21} suggesting UAE practice is comparable to other healthcare systems.

Interpretation in relation to guidelines

International guidance converges on prompt clozapine initiation once TRS is established. The International Guidelines for the Algorithmic Treatment of Schizophrenia guidelines recommend clozapine as early as 12 weeks after two unsuccessful trials, ^{Reference McCutcheon, Pillinger, Varvari, Halstead, Ayinde and Crossley7} aligning with National Institute for Health and Care Excellence guidance and other major guideline standards. ^{Reference Howes, McCutcheon, Agid, De Bartolomeis, Van Beveren and Birnbaum1,3,Reference McCutcheon, Pillinger, Varvari, Halstead, Ayinde and Crossley7,Reference Remington, Addington, Honer, Ismail, Raedler and Teehan18} Against this benchmark, most eligible patients in our cohort did not receive guideline-recommended treatment. It is important to acknowledge scientific nuance: a recent individual-participant data meta-analysis highlighted heterogeneity in randomised controlled trial evidence comparing clozapine with other second-generation antipsychotics. ^{Reference Schneider-Thoma, Hamza, Chalkou, Siafis, Dong and Bighelli22} Nevertheless, long-term observational studies consistently demonstrate clozapine’s superior effectiveness for mortality, hospital admission and suicide prevention. ^{Reference Taipale, Tanskanen, Mehtälä, Vattulainen, Correll and Tiihonen12,Reference Hatano, Kamei, Takeuchi, Gomi, Sakakibara and Hotta13}

Predictors and consequences of delayed initiation

A key contribution is the clear demonstration that clozapine is systematically positioned late in care pathways. ^{Reference Jakobsen, Schaug, Storebø, Austin, Nielsen and Simonsen23} Three converging findings support this conclusion. First, there was a strong dose–response relationship between number of prior antipsychotic trials and clozapine initiation probability (Spearman’s ρ = 1.00), with rates increasing from 4.3% among patients with two trials to 43.8% among those with six or more trials, and each additional adequate trial nearly doubling the odds of initiation (adjusted odds ratio 1.98, 95% CI 1.70–2.30; p < 0.001). Second, only a third of initiators received clozapine after exactly two failed trials as guidelines recommend; the majority (67.4%) required three or more, representing months of potentially avoidable ineffective treatment. Third, among delayed initiators, each additional adequate trial was associated with a longer delay to clozapine (+63.7% per trial; p < 0.001). This pattern of repeated switching, dose optimisation, augmentation and polypharmacy before clozapine consideration aligns with international literature describing ‘clozapine hesitation’. ^{Reference Howes, Vergunst, Gee, McGuire, Kapur and Taylor9} Under this strict definition, a substantial minority of initiators (53 out of 129) received clozapine before formally completing two adequate trials, indicating that clinical recognition of treatment resistance frequently precedes the formal threshold.

Beyond treatment sequencing, diagnostic factors may also contribute to delayed initiation. The lower prevalence of comorbid depression and bipolar disorder among initiators (10.9 v. 20.2% and 17.8 v. 28.1%, respectively) may reflect diagnostic gatekeeping, with clinicians reserving clozapine for ‘pure’ schizophrenia presentations despite evidence supporting its efficacy across schizophrenia spectrum disorders, including schizoaffective disorder. ^{Reference Wagner, Siafis, Fernando, Falkai, Honer and Röh24} Prior long-acting injectable exposure, by contrast, was not associated with clozapine initiation (adjusted odds ratio 0.92), subsequent hospital admission (adjusted incidence rate ratio 1.11) or the timing of initiation; we therefore do not interpret injectable use as a determinant of clozapine pathways in this cohort.

There are several possible explanations for the association between delay and hospital admission. The most straightforward explanation is that prolonged untreated TRS leads to ongoing symptomatic burden driving psychiatric crises, with delayed clozapine access allowing accumulation of relapse cycles, which may be associated with neurobiological changes, social disruption and progressive functional decline. ^{Reference Shah, Iwata, Plitman, Brown, Caravaggio and Kim25} However, confounding by severity represents an important alternative: patients with more severe presentations may face greater barriers to initiation while simultaneously having higher baseline hospital admission risk. ^{Reference Hatano, Kamei, Takeuchi, Gomi, Sakakibara and Hotta13} The non-monotonic pattern observed (highest incidence rate ratio at 12–24 months, lower at >24 months) may reflect selection effects, whereby patients experiencing very long delays represent a clinically distinct population from those with moderate delays. Our rate-adjusted approach addresses differential follow-up time, but cannot fully resolve confounding by indication. The association between delay and hospital admission persisted after adjustment for measured covariates; an exploratory propensity-score-weighted comparison was directionally consistent, but did not reach significance and was underpowered, so residual confounding cannot be excluded. Future studies with richer baseline data or instrumental variable approaches could further clarify causality.

UAE and Gulf context

The UAE population comprises both nationals (approximately 11.5%) and a large expatriate workforce from over 200 countries with employment-linked residency. ^{Reference Albanna, Sultan, Nawaz, Derby, Zafar Ali, Khan and Al-Shamsi26} Population mobility creates uncertainty about long-term follow-up availability, potentially reducing clinician willingness to initiate a medication requiring regular haematological monitoring, whereas high healthcare workforce turnover and heterogeneous training backgrounds may contribute to variation in prescribing practices. ^{Reference Paulo, Loney and Lapão27} Cultural factors add complexity: in Arab and Islamic cultures, collective community values are prioritised over individual autonomy, with family as the central unit influencing decision-making. ^{Reference Alageel, Alsadhan, Alkhaldi, AlKasabi and Alomair28} Language and cultural barriers have been shown to interfere with timely mental health diagnosis and impede development of therapeutic relationships necessary for complex treatment decisions, ^{Reference Alageel, Alsadhan, Alkhaldi, AlKasabi and Alomair28} and family involvement may prolong decision-making when multiple stakeholders require agreement. ^{Reference Khatib, Alyafei and Shaikh29} Cultural and societal stigma surrounding mental illness remains a significant barrier to treatment engagement in Gulf populations. ^{Reference Alageel, Alsadhan, Alkhaldi, AlKasabi and Alomair28}

Our nationality-stratified analyses allowed empirical testing of several hypothesised barriers. Initiation rates were comparable among expatriates and UAE nationals (7.7 v. 7.1%; p = 0.72), with no significant difference across nationality groups (p = 0.30). The absence of systematically lower initiation among expatriates suggests that clinicians may not be withholding clozapine based on perceived residence instability, or that other factors compensate for any such effect. ^{Reference Jakobsen, Schaug, Storebø, Austin, Nielsen and Simonsen23} Regional studies confirm structural and attitudinal barriers: a qualitative study across Arabian Gulf countries found monitoring burden among the most frequently cited deterrents, ^{Reference Ismail, Tounsi, Zolezzi and Eltorki30} only 53% of psychiatrists in the Middle East and North Africa reported using clozapine as first-line treatment for TRS, ^{Reference El Hayek, Noufi and Beayno31} and a systematic review of nine Arab countries confirmed consistently low prescribing rates. ^{Reference Al-Huseini, Qutishat, Khan and Ahmed14} Our quantitative findings suggest barriers are predominantly system-level, with implications for intervention design favouring universal process improvements rather than population-targeted strategies.

Concomitant mood-stabiliser and antidepressant use, although common over the treatment course, was substantially lower when measured cross-sectionally at the point of eligibility (7 and 10%, respectively) and was approximately twice as frequent in treatment-resistant patients with schizophrenia as in non-resistant patients with schizophrenia, consistent with greater clinical complexity rather than routine polypharmacy (Supplementary Fig. 7). A systematic review of the Arab world similarly documents consistent clozapine underutilisation driven by systemic and training-related barriers. ^{Reference Al-Huseini, Qutishat, Khan and Ahmed14} That international utilisation is unrelated to monitoring stringency ^{Reference Fitzgerald, O’Dwyer, Dhubhlaing, Gee, Sahm and Wheeler8} reinforces our inference that barriers here are predominantly system-level.

Temporal patterns and the era effect

Initiation rates calculated by year of treatment-resistance eligibility are confounded by differential follow-up, because patients becoming eligible in recent years have had less time to initiate clozapine, and are therefore not interpreted as a secular trend. Reported temporal trends elsewhere are inconsistent: Taiwan showed increasing rates over 12 years, ^{Reference Lin, Chan, Hsu and Chen32} whereas other studies found no improvement despite guideline revisions. A Danish registry study paradoxically found longer time-to-clozapine among patients diagnosed in 2003 compared with 1996, potentially reflecting the proliferation of alternative antipsychotics. ^{Reference Nielsen, Røge, Schjerning, Sørensen and Taylor20}

More encouraging was the finding that, among initiators, those becoming treatment-resistance-eligible in 2021–2025 had a shorter median delay than those eligible in 2018–2020 (11.5 v. 21.6 months, 46.9% shorter; p = 0.012) (Supplementary Fig. 8). This may reflect increased awareness through education and guideline dissemination, cumulative service learning or faster escalation for severe presentations. Similar reductions have been observed internationally, ^{Reference John, Ko and Dominic19} although findings are inconsistent across settings.

Era was not associated with the cumulative incidence of clozapine initiation (Fig. 4(b); log-rank p = 0.85). In models additionally including era of eligibility, era predicted neither the odds of ever initiating clozapine (adjusted odds ratio 1.03; p = 0.88) nor, in a Cox model accounting for differential follow-up, the initiation hazard (hazard ratio 1.37; p = 0.23; Supplementary Table 8), confirming no true between-era difference in the rate of initiation. The shorter delays in the recent era therefore reflect faster initiation among those who did initiate rather than a higher overall rate (Supplementary Fig. 8). ^{Reference John, Ko and Dominic19} The absence of secular improvement in initiation rates suggests passive guideline dissemination is insufficient. Active implementation through dedicated clozapine clinics and structured pathways has shown promise elsewhere, ^{Reference Barnes, MacCabe, Kane, Delgado, Khan and Paton33,Reference O’Donoghue, Mora, Bismark, Thompson and McGorry34} and formalising such approaches could accelerate progress locally.

Clinical and health service implications

The findings point to several actionable interventions. First, embed automatic eligibility triggers and decision support. Once two adequate antipsychotic trials are documented in the electronic medical record, automated prompts should flag treatment resistance and prompt structured consideration of clozapine. ^{Reference Correll, Agid, Crespo-Facorro, De Bartolomeis, Fagiolini and Seppälä35} This shifts the default from ‘clozapine as discretionary last resort’ to ‘clozapine as standard next step requiring documented opt-out’. Electronic clinical decision support has been successfully implemented for other high-stakes prescribing decisions and could be adapted for TRS identification. The prompt should trigger when the second adequate trial is documented as ineffective, consistent with the International Guidelines for the Algorithmic Treatment of Schizophrenia guidance recommending clozapine consideration as early as 12 weeks after two unsuccessful trials. ^{Reference McCutcheon, Pillinger, Varvari, Halstead, Ayinde and Crossley7}

Second, establish a dedicated clozapine pathway and clinic. International experience demonstrates that dedicated services with protected staffing and streamlined workflows can substantially improve access. ^{Reference Kelly, Freudenreich, Sayer and Love36} A dedicated clinic should include standardised baseline work-up, pre-booked phlebotomy appointments, pharmacist-led titration protocols and clear communication pathways with referring clinicians. ^{Reference Wagner, Siskind, Falkai, Howes, Correll and Lee37}

Third, reduce monitoring friction through logistics optimisation. Practical interventions include coordinated appointment bundles, point-of-care testing for immediate results ^{Reference Kamhi-Nesher, Taub, Halimi, Frenkel, Azam and Bormant38} and home phlebotomy services. ^{Reference Heins39} A pharmacy-based monitoring programme in Minnesota demonstrated that mobile phlebotomy increased programme uptake from 23 to 69% and extended mean treatment duration. ^{Reference Heins39} Recent consensus guidelines recommend reduced monitoring frequency after stable treatment, including 6-monthly monitoring after 2 years without neutropenia events. ^{Reference Siskind, Northwood, Pillinger, Chan, Correll and Cotes40}

Fourth, address workforce training heterogeneity. Given the multinational psychiatric workforce, mandatory competency-based training on clozapine initiation, titration and adverse effect management should be implemented. Practical evidence-based checklists for identifying and managing TRS can increase prescriber confidence and reduce delays. ^{Reference Correll, Agid, Crespo-Facorro, De Bartolomeis, Fagiolini and Seppälä35} Training should include management of common adverse effects and recognition of rare complications.

Fifth, implement interpreter and family engagement protocols. Collaboration with patients and families is essential but often overlooked. ^{Reference Alageel, Alsadhan, Alkhaldi, AlKasabi and Alomair28} For patients with limited English proficiency, standardised interpreter services and written information in Arabic, Hindi, Urdu, Tagalog and other commonly spoken languages should be available.

Strengths and limitations

This study has several methodological strengths. The large sample size (N = 1738 patients with TRS) represents the largest clozapine utilisation study from the Gulf region and provided adequate statistical power to examine multiple predictors simultaneously. Time-to-event methods appropriately handled censoring, and Poisson regression with robust standard errors and a person-time offset accounted for differential follow-up in outcome analyses. The operationalised TRS definition enhances international comparability, whereas objective prescription ascertainment from EMR data avoided recall bias. The dose–response relationship between treatment trials and initiation probability (Spearman’s ρ = 1.00) strengthens causal inference. Propensity-score adjustment partly explored potential confounding by indication in outcome analyses. Sensitivity analyses varying the required number of adequate trials, minimum trial duration and dose-adequacy threshold consistently demonstrated low initiation rates (between 6.5 and 14.1%), confirming robustness across definitional thresholds (Supplementary Table 2).

This study has several limitations. TRS was inferred from treatment exposure rather than prospectively confirmed non-response, potentially misclassifying patients whose medication changes reflected tolerability concerns. This would tend to overestimate TRS prevalence and underestimate the true clozapine initiation rate. We could not ascertain reasons for clozapine non-initiation, which may include contraindications, patient refusal or clinician preference. Associations between delay and outcomes may be confounded by unmeasured illness severity, which the exploratory, underpowered propensity-score analysis could not fully resolve. Our primary definition required a 6-week minimum trial duration, consistent with consensus guidelines; ^{Reference Howes, McCutcheon, Agid, De Bartolomeis, Van Beveren and Birnbaum1} a 4-week threshold was additionally examined in sensitivity analyses and yielded consistent findings (Supplementary Table 2). The 3-month concordance threshold reflects recent recommendations, ^{Reference McCutcheon, Pillinger, Varvari, Halstead, Ayinde and Crossley7} but is not universally specified. Because the data-set ended on 22 December 2025, patients who became treatment-resistant shortly before data extraction may not have had the full 3-month observation window to demonstrate guideline-concordant initiation; Kaplan–Meier censoring partly addresses this, but concordance estimates should be interpreted with this right-censoring limitation in mind. EMR availability from 2018 introduces potential left truncation bias, although restricting the cohort to patients whose first recorded contact occurred after 2018 yielded a similar initiation rate (7.2 v. 7.4%; Supplementary Table 2). Finally, this single-centre study from a specialised psychiatric hospital in Dubai may not generalise to other settings or regions with different prescribing cultures.

In conclusion, this study adds regional evidence from an underrepresented healthcare context to a literature dominated by European, North American and Australian data. In this large UAE cohort, clozapine was initiated in only 7.4% of TRS-eligible patients, with cumulative initiation of approximately 10% at 7 years. Median delay was 13.2 months; only a quarter achieved guideline-concordant initiation. Consistent with international studies, clozapine was systematically positioned late in treatment pathways, with a clear dose–response relationship between prior trials and initiation probability. Delayed initiation was associated with higher hospital admission rates, an association robust to adjustment for measured covariates, although an exploratory propensity-score analysis was underpowered. Importantly, barriers appeared system-wide: nationality did not significantly predict initiation, suggesting interventions should target universal process improvements rather than specific populations. These findings establish a UAE baseline for quality improvement and support systematic efforts to improve timely access to clozapine for patients meeting treatment-resistance criteria.