3.1 Demographics and Clinical Characteristics of the Whole Cohort (with and without SUD)

The combined with SUD and without SUD sample consisted of 471 participants, including 143 people with SUD (83.2% males, 28.86 ± 9.13 years old; illness duration 2.87 ± 1.74 years; PANSS total score at baseline = 72.20 ± 19.44; combined LAI antipsychotic arm N = 74, combined oral antipsychotics arm N = 69) and 328 without SUD (59.5% males, 31.42 ± 9.86 years old; illness duration 3.09 ± 1.91 years; PANSS total at baseline = 75.69 ± 18.17; combined LAI antipsychotic arm N = 174, combined oral antipsychotics arm N = 154). Table 1 provides additional details on the clinical and demographic characteristics of the whole cohort.

Table 1 Demographics and clinical characteristics of individuals included in this secondary analysis of the European Long-Acting Antipsychotics in Schizophrenia Trial (EULAST)

Independent t-tests showed no difference between the two groups regarding PANSS total at baseline (t = 1.78, degrees of freedom (df) = 454, P = 0.08, 95% confidence intervals CI −0.37 to 7.34) and illness duration (t = 1.19, df = 469, P = 0.23, 95% CI − 0.14 to 0.57). Moreover, the SUD sample was to be marginally younger than the without SUD sample (28.86 ± 9.13 years old versus 31.42 ± 9.86 years old, respectively; t = 2.73, df = 469, P = 0.007, 95% CI 0.72–4.41). The chi-squared test of independence showed a significantly higher male/female proportion in substance users than in not substance users (χ2 = 24.25, df = 1, P < 0.001).

3.2 Demographics and Clinical Characteristics of SUD (oral versus LAI)

The mean age of the SUD population was 28.86 ± 9.13 years, and 16.8% were female. The mean duration of illness was 2.87 ± 1.74 years. Among the participants with SUD, N = 73/143 (51%) received aripiprazole and 70/143 (49%) received paliperidone. In the oral versus LAI groups there were no significant differences regarding any clinical parameter including age (Z = − 1.081, P = 0.280), duration of illness (Z = − 0.503, P = 0.615), sex (χ2 = 0.404, P = 0.525), aripiprazole versus paliperidone (χ2 = 0.168, P =0.682), and PANSS total scores at baseline (Z = − 0.882, P = 0.378). Table 2 provides additional details on the clinical and demographical characteristics of the SUD cohort.

Table 2. Demographics and clinical characteristics of individuals with substance use disorder (SUD)3.3 Demographics and Clinical Characteristics of without SUD (oral versus LAI)

The mean age of the without SUD population was 31.42 ± 9.86 years, and 40.5% were female. The mean duration of illness was 3.09 ± 1.91 years. N = 158/328 (48.2%) received aripiprazole, and 170/328 (51.8%) received paliperidone. In the oral versus LAI groups, there were no significant differences regarding any clinical parameter including age (Z = − 0.388, P = 0.698), duration of illness (Z = − 0.915, P =0.360), sex (χ2 = 2.815, P = 0.093), aripiprazole versus paliperidone (χ2 = 0.389, P = 0.533), and PANSS total scores at baseline (Z = − 0.88, P = 0.38). Supplementary Table 1 provides additional details on the clinical and demographical characteristics of the without SUD cohort.

3.4 Lower Hazard Ratios for LAIs than Oral Antipsychotics in SUD

The Cox proportional hazard regression model in SUD demonstrated moderate concordance of 0.595 (standard error, SE = 0.028), with a statistically significant fit as supported by the likelihood ratio test (χ2 (3) = 11.88, P = 0.008) and Wald test (χ2 (3) = 11.43, P = 0.01).

The model indicated a significant effect for mode of antipsychotic treatment (LAI versus oral) on adjusted hazard ratios (HRs) of all-cause discontinuation in SUD (adjusted HR = 0.641, 95% CI 0.438–0.938, P = 0.022). Kaplan–Meier curves (Fig. 2a) demonstrated favorable survival trajectories for LAI antipsychotics (median time to all-cause discontinuation = 158 days, events = 50, censored = 24) than oral antipsychotics (median time to all-cause discontinuation days = 97, events = 59, censored = 10). Moreover, older age in participants was significantly associated with lower hazard ratios of all-cause discontinuation (adjusted HR = 0.973, 95% CI 0.952–0.995, P = 0.015). An interaction factor between age and mode of antipsychotic treatment was incorporated in the model to examine the modulatory impact of these two significant factors on each other. The interaction term did not yield a significant effect on the hazard ratio (P > 0.1), yet the likelihood ratio and Wald test continued to demonstrate statistical significance for the model following the introduction of the interaction term. This suggests a similar modulation of hazard ratios by age in LAI and oral antipsychotic treatments. The model revealed no significant impact for sex on the hazard ratios (HR = 0.755, 95% CI 0.463–1.231, P = 0.260).

Fig. 2

Treatment discontinuation of LAI and oral antipsychotic treatments because of any cause in participants (A) with comorbid SUD and in (B) without comorbid SUD. Dashed lines indicate median time without treatment discontinuation under treatment. P values for the effect of mode of antipsychotic treatment (oral versus LAI) on the adjusted HR of all-cause discontinuation are shown in each plot. LAI long-acting injectable

The model examining the crude hazard ratio of all-cause discontinuation in SUD sample revealed similar findings, showing a significant effect for mode of antipsychotic treatment (LAI versus oral; crude HR = 0.659, 95% CI 0.451–0.962, P = 0.031) with favorable survival rates for LAI antipsychotics, with a statistical significance for the overall model according to likelihood ratio (χ2 (1) = 4.7, P = 0.03) and Wald test (χ2 (1) = 4.74, P = 0.03).

3.5 Lower Hazard Ratios for LAIs than Oral Antipsychotics in Current SUD

The Cox proportional hazard regression model including subjects with current SUD had an overall model fit similar to that of the extended SUD sample, with a moderate concordance of 0.54 (SE = 0.031) and a chi-squared coefficient of 3.51 (df = 3) and 3.48 (df = 3) in the likelihood ratio and Wald test, respectively. Both tests showed, however, a marginally significant fit (P = 0.06), most probably due to a low statistical power of the subsample and a resulting higher type II error. The model showed lower crude-hazard ratios of all-cause discontinuation for LAI compared with oral antipsychotics for current substance users (crude HR = 0.65, 95% CI 0.42–1.02 P = 0.062), which were statistically marginally significant. These results replicate the statistically significant lower crude and adjusted HRs of all-cause discontinuation observed in the extended SUD sample.

3.6 Comparable Hazard Ratios for LAIs and Oral Antipsychotics in Non-SUD

The Cox proportional hazard regression model in non-SUD demonstrated a poor discriminative performance with a concordance index comparable to chance (0.52, SE = 0.021). The likelihood test (χ2 (3) = 2.38, P = 0.5) and Wald test (χ2 (3) = 2.37, P = 0.5) revealed that the model is not a statistically significant predictor of hazard ratios of all-cause discontinuation. Consequently, none of the factors within the model were found to be significantly associated with hazard ratios of all-cause discontinuation. This included mode of antipsychotic treatment (LAI versus oral; P = 0.282), age (P = 0.69), and sex (P = 0.4). Kaplan–Meier curves (Fig. 2b) demonstrated comparable survival trajectories for LAI antipsychotics and oral antipsychotics in non-SUD.

Similar findings were found by the model examining the crude hazard ratio for all-cause discontinuation in the non-SUD sample. The overall model was not statistically significant as per likelihood ratio test (χ2 (1) = 1.27, P = 0.3) and Wald test (χ2 (1) = 1.28, P = 0.3), with a nonsignificant effect for mode of antipsychotic treatment (LAI versus oral; crude hazard ratio = 0.855, 95% CI 0.651–1.122, P = 0.259).