Long-term trends in age at psychiatric admission (2006–2022): influence of sex, diagnosis, and admission type in a single Italian catchment

Abstract

Aims

Following Italy’s psychiatric reform, national inpatient numbers declined, but how the age at admission has changed across sex, diagnostic, and admission-type subgroups remains unclear. This study examined 17 consecutive years (2006–2022) of psychiatric admissions to the sole ward serving L’Aquila, assessing temporal trends in age at admission by diagnosis, sex, and admission status (voluntary vs. compulsory).

Methods

All adult admissions (≥ 18 years) were extracted from hospital discharge records (Schede di Dimissione Ospedaliera, SDO). Analyses were conducted at the admission-episode level. Primary diagnoses were grouped into four ICD-9 categories: schizophrenia spectrum, major depressive disorder, bipolar disorder, and alcohol/substance use disorder. Multiple linear models tested time-related changes in mean age at admission, including interactions for time × diagnosis × sex × admission type.

Results

Across 5,207 admission episodes, the Trimester × Diagnosis interaction showed a marked decline in age at admission for major depression (B = − 0.20 per trimester, p < 0.001) and bipolar disorder (B = − 0.09, p = 0.03), equivalent to approximately − 0.80 and − 0.36 years per year, yielding total reductions of ~ 13.6 and ~ 6.1 years over 2006–2022. The Trimester × Diagnosis × Sex interaction indicated that the decline in depression was driven by men (B = − 0.32, p < 0.001; ≈ −1.28 years/year; −21.76 over 17 years), while in bipolar disorder it was driven by women (B = − 0.14, p = 0.03; ≈ −0.56 years/year; −9.52 over 17 years). The Trimester × Admission type interaction showed the reduction was specific to voluntary admissions (B = − 0.10, p < 0.001; ≈ −0.40 years/year), while compulsory admissions were stable (B = − 0.01, p = 0.87). No significant age change occurred in schizophrenia spectrum disorders or in alcohol/substance use disorders.

Conclusions

Between 2006 and 2022, we observed a progressive rejuvenation of voluntary and affective-disorder inpatient admissions, with the clearest subgroup-specific reductions detected in men with major depression and women with bipolar disorder, while remaining stable in schizophrenia spectrum, alcohol/substance use disorder, and compulsory admissions. These findings underscore the need to balance youth-focused outreach with adequate capacity for chronic psychosis and substance use disorder.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12991-026-00681-2.

Introduction

The Italian psychiatric reform (Law 180) closed asylums and relocated acute psychiatric beds to general hospital wards, establishing a community-oriented model that reduced institutionalization while preserving an inpatient safety net for clinical crises [1–5].

While national hospitalization rates have largely stabilized since the mid-2000s, far less is known about how age at admission has evolved within local catchments. International and Italian evidence suggests that reorganization of services and early-intervention models may shift the demographic composition of inpatient cohorts toward younger, help-seeking populations, particularly for affective disorders [6–12]. In Italy, early post-reform investigations of general-hospital psychiatric wards already described age-at-first-admission patterns and sex differences, indicating that women tend to be older than men at first hospital contact and that local system configuration shapes the profiles reaching inpatient care [13, 14]. Multicenter surveys then provided concrete age benchmarks for acute wards: the PERSEO study across 62 SPDCs reported a mean age at admission in the mid-forties with women older than men, and complementary national work on first-ever inpatients found a median admission age around 40 years, consistent with a younger profile at incident admission compared with the prevalent inpatient population [15, 16].

Globally, most mental disorders begin before age 25, as shown by large-scale epidemiological studies [17]. This makes earlier detection particularly plausible for conditions characterized by preserved insight, such as major depression and bipolar disorder. Conversely, disorders marked by impaired insight—such as psychotic disorders—tend to present in late adolescence or early adulthood; meta-analytic data indicate their median age at first hospitalization is around 24–25 years, reflecting relatively stable age-at-admission distributions across cohorts [18]. Substance use disorder–related hospital contacts may show heterogeneous, age-band–specific changes rather than a uniform linear shift in mean age; for example, an analysis of hospital discharge registries from the Abruzzo region (2006–2015) reported differential temporal patterns across age ranges in drug-dependence hospitalizations [19].

Recent observational evidence—albeit in younger settings—likewise reports no age shift for schizophrenia but declining age for affective disorders, consistent with diagnosis-specific trajectories in help-seeking and referral [20]. In the present study, we analyzed seventeen consecutive years (2006–2022) of admissions to the sole 15-bed psychiatric ward serving L’Aquila, focusing exclusively on temporal trends in age at admission by diagnosis, sex, and admission status (voluntary vs. compulsory), to provide high-resolution local evidence for service planning.

Methods

Participants and data extraction

We retrieved all discharge summaries (Scheda di Dimissione Ospedaliera, SDO) from the psychiatric ward of the San Salvatore Hospital in L’Aquila (Italy) covering the period 2006–2022.

The data extraction, quality checks, and analytic workflow followed the approach described in our previous report on admission patterns in the same ward [21].

Hospital discharge records included sociodemographic information, admission and discharge dates, and the primary diagnosis at discharge, coded according to the International Classification of Diseases, Ninth Revision (ICD-9). ICD-9 codes used to define each diagnostic category are provided in Supplementary Table S2.

ICD-9 was employed because the regional health information system of Abruzzo is still based on this classification, which is used across the electronic medical record systems, archives, and discharge documentation. To preserve privacy, all patient identifiers were anonymized prior to analysis. Because discharge records were anonymized and did not include a stable unique patient identifier for longitudinal linkage, the unit of analysis was the admission episode. Repeated admissions from the same individual could not be identified and were therefore analyzed as separate episodes. Sex was extracted from administrative hospital discharge records as recorded at admission (male/female).

After verifying data integrity, completeness, and internal consistency, all variables were checked for missing or inconsistent values. Episodes with missing data on key variables (age, trimester, primary diagnosis, or admission type) were excluded via complete-case analysis (n = 503), yielding a final analytic sample of 5,207 admission episodes. Participants were grouped into four diagnostic categories, based on the ICD-9 discharge diagnosis: schizophrenia spectrum disorders, major depressive disorder, bipolar disorder, and alcohol/substance use disorder.

Statistical analysis

All analyses were performed in R (version 4.1.2, R Foundation for Statistical Computing, Vienna, Austria). The date of admission was used to assign each hospitalization to its corresponding trimester, resulting in 68 consecutive quarters covering the entire 17-year observation window (2006–2022). The quarter variable was mean-centered to facilitate interpretation of interaction terms and reduce multicollinearity; the centering constant was retained to reconstruct the original time scale for graphical representation.

Descriptive statistics were computed to summarize demographic and clinical characteristics across diagnostic groups. Continuous variables (e.g., age) are reported as mean ± standard deviation (SD), while categorical variables (sex: male/female, admission type: voluntary/compulsory) are reported as absolute and relative frequencies (N, %).

To assess whether age at admission varied as a function of time, diagnosis, sex, and admission type, we fitted a multiple linear model including all possible interactions up to the four-way level (time × diagnosis × sex × admission type). Sum-to-zero contrasts were applied to categorical factors, and Type-III ANOVA tables were computed using the car package. In the presence of significant main effects or interactions, simple slope analyses were conducted (emmeans, jtools) to clarify the source of effects. All statistical tests were two-tailed, with the significance threshold set at α = 0.05.

Model assumptions and diagnostics were formally evaluated. Residual autocorrelation was not detected (Durbin–Watson = 1.97, p = 0.16). A Breusch–Pagan test indicated heteroskedasticity (BP = 63.51, p < 0.001). The residual distribution deviated from normality on a large-N Shapiro–Wilk test (W = 0.99, p < 0.001), a common and negligible issue in ordinary least squares (OLS) models with > 5,000 observations due to the asymptotic normality of estimators. Influence diagnostics (Cook’s distance and leverage–residual plots) did not reveal influential observations of concern. Multicollinearity was modest: for all predictors and interactions, GVIF^1/2ᴰᶠ ≤ 2.07, well below conventional alert thresholds. To address the detected heteroskedasticity, all tests were re-estimated with heteroskedasticity-consistent standard errors (HC3). The pattern of statistical significance remained unchanged across main effects and interactions. Given the stability of results using robust estimators, the absence of autocorrelation, the lack of influential outliers, and the need for full compatibility with the subsequent simple-effects decomposition, we report OLS estimates with conventional standard errors in the main text, while providing robust (HC3) results as a supplementary robustness check (see Supplementary Table S1).

Results

A total of 5,710 discharge records were retrieved; 503 episodes were excluded due to missing data on age, trimester, diagnosis, or admission type, resulting in 5,207 admission episodes included in the analyses.

Demographic and admission characteristics of the overall sample are reported in Table 1, while model results are summarized in Table 2.

Table 1.

Demographic and admission characteristics of psychiatric admission episodes recorded between 2006 and 2022, stratified by discharge diagnosis

Diagnosis	Age mean ± SD	Sex N (%)		Admission type N (%)
		Male	Female	Compulsory	Voluntary
Schizophrenia spectrum	44.2 ± 14.47	1649 (57.4)	1224 (42.6)	493 (17.16)	2380 (82.84)
Major depression	49.57 ± 16.29	408 (44.88)	501 (55.12)	116 (12.76)	793 (87.24)
Bipolar disorder	47.23 ± 15.08	468 (49.63)	475 (50.37)	145 (15.38)	798 (84.62)
Alcohol/substance use disorder	43.72 ± 14	324 (67.22)	158 (32.78)	60 (12.45)	422 (87.55)
Total	45.64 ± 15.02	2849 (54.71)	2358 (45.29)	814 (15.63)	4393 (84.37)

Table 2.

Results for the linear model predicting age at admission as a function of trimester, diagnosis, sex, and admission type, including all possible interactions

Effect	F-value	p -value
Trimester	5.35	0.02
Diagnosis	14.75	< 0.001
Sex	7.74	0.005
Admission type	6.66	0.01
Trimester × Diagnosis	7.16	< 0.001
Trimester × Sex	0.30	0.59
Diagnosis × Sex	1.59	0.19
Trimester × Admission type	4.01	0.04
Diagnosis × Admission type	0.55	0.65
Sex × Admission type	1.45	0.23
Trimester × Diagnosis × Sex	2.77	0.04
Trimester × Diagnosis × Admission type	0.96	0.41
Trimester × Sex × Admission type	2.06	0.15
Diagnosis × Sex × Admission type	0.56	0.64
Trimester × Diagnosis × Sex × Admission type	0.72	0.54

Significant main effects and interactions are highlighted in bold

The model revealed significant main effects of Trimester, Diagnosis, Sex, and Admission type. Age at admission therefore varied across diagnostic groups, sexes, and admission types, and showed a temporal reduction over the 17-year period.

Among two-way interactions, the Trimester × Diagnosis and Trimester × Admission type were significant, indicating that temporal variations in admission age differed across diagnostic and admission-type subgroups. Moreover, a significant three-way interaction (Trimester × Diagnosis × Sex) indicated that the differential temporal trends in admission age across diagnostic categories were further modulated by patients’ sex. No other three-way or four-way interactions reached significance (p ≥ 0.15).

Simple slope analyses on Trimester × Diagnosis interaction (Fig. 1) showed that mean age at admission significantly decreased across years for inpatient admissions with major depression (B = − 0.20, p < 0.001) and bipolar disorder (B = − 0.09, p = 0.03), while remaining stable for schizophrenia spectrum (B = 0.02, p = 0.26) and alcohol/substance use disorders (B = 0.06, p = 0.33). These results indicate a progressive reduction in the average age of affective-disorder admissions, corresponding to an estimated decrease of approximately 0.80 years per year for depression and 0.36 years per year for bipolar disorder. Over the entire observation window (2006–2022), this translates into a total reduction of about 13.6 years in admission age for depression and 6.1 years for bipolar disorder.

Fig. 1.

Predicted admission age trajectories by diagnosis. Lines represent model-estimated marginal means, and shaded areas indicate the standard error of the mean. Asterisks denote significant temporal effects for the corresponding color-coded group (*p < 0.05, ***p < 0.001)

The Trimester × Diagnosis × Sex interaction (Fig. 2) clarified that these temporal effects were sex-specific: in major depression, the decline in admission age was mainly driven by male patients (B = − 0.32, p < 0.001), whereas in bipolar disorder the decreasing trend was primarily explained by female admissions (B = − 0.14, p = 0.03). These slopes correspond to a mean annual decrease of 1.28 years per year for men with depression and 0.56 years per year for women with bipolar disorder, amounting to a total reduction of 21.76 years and 9.52 years, respectively, over the 17-year observation period. No significant changes over time were observed in the opposite sex within each diagnostic group, nor for schizophrenia or alcohol/substance use disorders (all p ≥ 0.10).

Fig. 2.

Predicted admission age trajectories by diagnosis and sex. Lines represent model-estimated marginal means, and shaded areas indicate the standard error of the mean. Asterisks denote significant temporal effects for the corresponding color-coded group (*p < 0.05, ***p < 0.001)

Finally, the Trimester × Admission type interaction (Fig. 3) revealed that the overall reduction in age at admission was specific to voluntary hospitalizations (B = − 0.10, p < 0.001), whereas compulsory admissions remained stable across the observation period (B = − 0.01, p = 0.87). This corresponds to a mean annual decrease of 0.40 years per year among voluntary admissions, totaling about 6.80 years over 2006–2022.

Fig. 3.

Predicted admission age trajectories by admission type. Lines represent model-estimated marginal means, and shaded areas indicate the standard error of the mean. Asterisks denote significant temporal effects for the corresponding color-coded group (*p < 0.05, ***p < 0.001)

Taken together, these results indicate a gradual younger age profile among affective-disorder admissions and voluntary admissions, with the clearest subgroup-specific decline observed in men with major depression and women with bipolar disorder, whereas schizophrenia spectrum, alcohol/substance use disorder, and compulsory admissions remained stable over time.

Discussion

This seventeen-year single-catchment analysis demonstrates a clear temporal shift in age at psychiatric admission. Age at admission declined steadily between 2006 and 2022, entirely driven by voluntary admissions and most marked among men with major depression and women with bipolar disorder, whereas compulsory, psychotic-disorder and alcohol/substance use disorder admissions remained stable.

Set against Italian post-reform benchmarks, these patterns are expected. In the national PERSEO survey across 62 SPDCs, the mean age at admission was in the mid-forties and women were older than men, offering a pragmatic reference for sex differences at entry to inpatient care; national data on first-ever inpatients reported a median admission age around 40 years, consistent with a younger profile at incident admission compared with the prevalent inpatient population [16, 22]. Early post-reform ward studies had already documented age-at-first-admission patterns and sex differences in neighboring catchments with different service configurations [13, 14].

Within this context, the decline in age at admission for affective disorders may reflect, at least in part, earlier recognition and help seeking among younger adults [47], consistent with rising mental health literacy and reduced stigma [23, 24, 44], and the diffusion of community and early intervention pathways [7–10, 25]. However, these interpretations remain indirect, as age at inpatient admission may also be influenced by unmeasured system-level factors such as bed availability, admission thresholds, emergency department referral patterns, and local service organization.

The fact that the reduction is confined to voluntary admissions is compatible with a process driven by awareness and autonomous engagement with care, but it may also reflect changes in service organization or admission practices over time [26,27].

By contrast, the stability of age at admission for schizophrenia accords with impaired insight and crisis-driven presentations that more often culminate in compulsory admission [28, 29]; in this scenario, stable age patterns are expected, as also suggested by cohort evidence reporting no age shift for schizophrenia while affective disorders show decreasing age [20]. Italian studies on coercion and psychosis further support a persistent, crisis-driven pathway for schizophrenia [30–32].

Sex-specific effects strengthen this reading. Among men with depression, the steep decline suggests an erosion of traditional barriers to seeking help; among women with bipolar disorder, a moderate but significant reduction points to improved recognition and timelier intervention [12]—both consistent with known gender differences in help-seeking [33–35, 46] and with the PERSEO observation that women present older at admission on average [22].

Importantly, age at admission for alcohol/substance use disorders remained stable across the observation window. This net stability may reflect the coexistence of different cohort dynamics—such as an ageing segment of long-term substance users alongside increasing visibility of alcohol- and drug-related problems in younger groups—that can offset each other when admissions are analyzed in aggregate[45]. Evidence on age- and cohort-specific trends in substance-use admissions and on changing alcohol-attributable morbidity supports the plausibility of these counterbalancing processes [36–39, 43]. Consistent with this interpretation, a discharge-registry study from the Abruzzo region documented age-range–specific temporal changes in drug-dependence hospitalizations, supporting the possibility that aggregate stability in mean age may coexist with divergent trends within age strata [19].Moreover, the absence of a temporal shift in mean admission age for alcohol/substance use disorders should not be equated with demographic stagnation. This broad category likely aggregates heterogeneous pathways (e.g., alcohol vs. other substances; primary SUD vs. dual diagnosis), in which opposite cohort dynamics may offset each other at the level of the mean. In addition, within the Italian service configuration, a substantial proportion of SUD care is delivered through addiction services, and psychiatric ward admissions may selectively capture crisis-driven or comorbidity-heavy presentations, which could sustain a relatively stable age profile over time. Future work should disentangle alcohol- vs. drug-related admissions and examine distributional changes (e.g., age quantiles or age-band-specific admission rates) to test whether stability in the mean masks divergent trends at the extremes.

Finally, the 2006–2022 observation window encompassed major exogenous shocks for this catchment, notably the 2009 L’Aquila earthquake and the COVID-19 pandemic. These events may have influenced psychiatric admissions through multiple pathways, including service reorganization, changes in access thresholds, psychosocial stressors, fear-related psychopathological burden, and delayed help-seeking. In particular, emerging Italian evidence suggests that the pandemic affected both the subjective psychiatric burden associated with COVID-19-related fear and broader patterns of mental health service utilization and psychiatric hospital admissions [21, 40–42]. Because time was modelled as a single linear effect across quarters, the estimated slopes should be interpreted as average trends over the whole period rather than as homogeneous changes across all subperiods. Future studies should test potential structural breaks or interrupted time-series effects to distinguish pre-event, acute-event, and post-event phases more directly.

Strengths of the study include the long observation window, complete catchment coverage, and a modelling strategy that tested higher-order interactions alongside comprehensive diagnostics and robust (HC3) checks.

Limitations

Several limitations should be considered when interpreting these findings. First, analyses were conducted at the admission-episode level rather than at the individual-patient level. Because the anonymized administrative dataset did not include a stable unique identifier, repeated admissions from the same individual could not be linked over time. Frequent service users may therefore have contributed multiple episodes, potentially influencing subgroup-specific age profiles and temporal slopes. Accordingly, the observed trends should be interpreted as changes in the age distribution of inpatient admission episodes, not as changes in the age of unique individuals entering inpatient care.

Second, diagnoses were based on the primary ICD-9 discharge diagnosis recorded in administrative data. Although this is standard practice in registry-based studies, it may have introduced some diagnostic misclassification and does not fully capture the clinical complexity of psychiatric inpatient presentations, including comorbid psychiatric disorders, substance use comorbidity, illness severity, or possible changes in coding practices over time.

Third, time was modelled as a linear effect across the entire 2006–2022 period. As discussed above, this approach summarizes the average long-term trajectory but may obscure non-linear changes associated with major contextual events, including the 2009 L’Aquila earthquake and the COVID-19 pandemic.

Fourth, the alcohol/substance use disorder category aggregated heterogeneous conditions, including alcohol-related disorders, illicit substance use, abuse versus dependence, and primary substance use versus dual-diagnosis presentations. This grouping may have masked distinct age-specific hospitalization trajectories within the broader category.

Finally, sex was recorded administratively as a binary variable and therefore does not capture gender identity or non-binary classifications. Nonetheless, the consistency of the results across model specifications and robustness checks supports the reliability of the overall pattern observed.

Conclusion

From 2006 to 2022, age at psychiatric admission declined over time among affective-disorder and voluntary admissions, with the clearest subgroup-specific reductions observed in men with major depression and women with bipolar disorder, while remaining stable in schizophrenia spectrum and alcohol/substance use disorders and among compulsory admissions. This pattern may reflect increased awareness and earlier, self-initiated help-seeking for conditions with better insight, alongside a persistent, crisis-driven pathway for schizophrenia spectrum and alcohol/substance use disorder. Service planning should strengthen youth- and gender-sensitive early-intervention and psychoeducation to support timely voluntary care, while maintaining high-dependency capacity and assertive follow-up for chronic psychosis. Ongoing, high-resolution local surveillance of age-at-admission trajectories will be essential to monitor these demographic shifts and guide balanced resource allocation.

Author contributions

TB, FP, FR and FS conceived and designed the project idea. TB, GR, FS, and ARacquired and the data. TB, AB, FP, FS, ST, GR and VS drafted the manuscript. ET, FS, ST, AB and AR drew figures and tables. All authors edited the draft and approvedthe final manuscript.

Funding

None.

Data availability

The datasets used and/or during the current study are available fromthe corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All procedures performed in studies involving human participants were conducted ethically, adhering to the institutional guidelines and the 1964 Declaration of Helsinki, including subsequent updates. The Ethics Committee obtained ethical approval from the Internal Review Board of the University of L’ Aquila on 18/10/2016, Sequence No: 05/2016. The participant consent requirement was waived because of the retrospective design, based on data from the records of the San Salvatore Hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Generative AI statement

The author(s) declare that Generative AI was used in the creation of this manuscript. The authors declare that they used ChatGPT (OpenAI) for language editing and proofreading. The authors confirm that they have verified the accuracy of all content, including citations, references, data, and accept full responsibility for the accuracy and integrity of the results presented.