Introduction
Frailty is characterized by a multisystem dysregulation of physiological reserve. When stressors arise—such as infection, surgery, or falls—older adults with diminished physiological reserve are at greater risk of adverse outcomes.1 Frailty reliably predicts falls, functional decline, rehospitalization, and death in older adults.1,2 Recent evidence has linked frailty to schizophrenia through shared genetic variants, brain structural alterations, and peripheral biomarkers.3 Given this potential shared biological basis, assessing frailty in patients with schizophrenia warrants clinical attention.
Schizophrenia is increasingly recognized as an accelerated-aging disorder, placing older patients at elevated risk of frailty.4,5 Consistent with this, physical decline occurs earlier and frailty is more prevalent in these patients than in age-matched peers.5,6 However, aging alone does not fully explain this phenomenon. Negative symptoms, cognitive impairment, and metabolic disturbances deplete physiological reserve; antipsychotic side effects such as sedation and extrapyramidal symptoms further contribute.7–9 The cumulative burden is greatest in those with decades of illness and long-term medication exposure.
Clinical evidence supports this association. Among older Chinese inpatients with schizophrenia, loss of independence in activities of daily living is linked to greater frailty, in part from poor nutrition.10 Frailty is also more common in treatment-resistant schizophrenia.11 Its key features — low physical activity and slow gait — reliably predict falls.12 These consequences are clinically significant: after hip fracture, people with schizophrenia are less likely to receive adequate care.13 Therefore, fall-related vulnerability warrants clinical attention in older adults with psychosis.
Most existing evidence on frailty risk comes from community-dwelling older adults, where advanced age, female sex, polypharmacy (particularly anticholinergic burden), chronic diseases, and malnutrition are well-established risk factors.14,15 However, it remains unclear whether these factors generalize to hospitalized patients with schizophrenia. Frailty in this population may arise from a combination of multiple clinical stressors. Polypharmacy and anticholinergic exposure accelerate muscle loss and functional decline,16,17 while common comorbidities, including diabetes and cardiovascular disease, heighten systemic inflammatory and metabolic stress.18 Additionally, anxiety, cognitive impairment, and sensory impairment may further deplete physiological reserve.19 Metabolomic studies have identified shared biological pathways linking frailty and late-onset schizophrenia.20
In contrast, frailty among hospitalized patients with schizophrenia remains understudied.4,21 One previous study attempted to predict frailty in psychiatric inpatients based on physiological, cognitive, and sensory indicators.22 Frailty is nonetheless highly prevalent in this population and closely linked to falls, poor fracture recovery, and increased mortality.1,12,13 The prevalence and risk factors of frailty in this specific population remain inadequately characterized. Although physical activity is recommended for older adults,23,24 effective targeting requires identifying those most vulnerable.25 We found no existing nomogram for frailty identification in this group using routine data. Therefore, we developed and internally validated a nomogram based on routinely collected clinical data to identify current frailty, requiring no specialized equipment.
This study was reported following the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis Or Diagnosis (TRIPOD) statement.
Materials and Methods
Study Design and Population
This cross-sectional study recruited older patients with schizophrenia using convenience sampling from the psychiatric unit of a tertiary Grade A hospital in Guangzhou between May and September 2025. Patients were eligible if they met the following criteria: (1) schizophrenia diagnosed by a psychiatrist according to the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5); (2) age ≥ 60 years; and (3) written informed consent from participants or their legal guardians for those with impaired decision-making capacity. Patients were excluded for any of the following reasons: (1) severe cognitive impairment (eg, moderate-to-severe dementia) that precluded cooperation with the assessment; or (2) concomitant severe physical illnesses (eg, acute myocardial infarction or stroke) or unstable medical conditions.
A total of 296 elderly inpatients with schizophrenia were screened. Among them, 17 patients were excluded due to refusal to sign the informed consent form or withdrawal during the study period. Eventually, 279 patients were included in the final analysis.
No missing data were present for any variable included in the final analysis. The sample size was estimated based on an anticipated maximum of 11 statistical variables (10 clinical variables, with sensory impairment coded as two dummy variables), requiring at least 110 events. Although the observed number of frail events was 106, slightly below the initially planned 110 events, the final model retained six statistical variables, yielding an EPV of 17.7, which exceeds the commonly recommended threshold of 10.
This indicates that the final model was not underpowered for the number of variables actually included. The study protocol was filed with the National Medical Research Registration and Filing Information System of China (Filing No: MR-44-25-064899). This study was conducted in accordance with the Declaration of Helsinki.
Ethical approval was obtained from the Institutional Review Board of The Affiliated Brain Hospital of Guangzhou Medical University (Approval No: 2025–034). Written informed consent was obtained from all participants or their legal guardians for those with impaired capacity.
Measures
Demographic and Clinical Data
We extracted the following data from electronic medical records and patient interviews: age, sex, education, body mass index (BMI), number of diagnoses, number of family visits in the past year, disease duration (years since first diagnosis), length of current hospitalization (years since current admission), smoking status (current smoker: ≥1 cigarette/day for ≥6 months), fall history (≥1 fall in the past year), family psychiatric history, number of medications (all oral and injectable medications), walking aid use (cane, walker, or wheelchair; hereafter referred to as walking aid), somatic pain (in the past month), surgical history, and swallowing function. Activities of daily living were assessed by the Barthel Index. Sensory impairment was extracted from the electronic medical records and categorized as none, single (visual or hearing impairment), or dual (both visual and hearing impairment).
Fried Frailty Phenotype
We applied the Fried frailty phenotype to define frailty.26 It comprises five components: (1) unintentional weight loss (>10 lb (4.5 kg) in the past year at baseline); (2) slow gait speed, defined as time to walk 4.57 m (15 feet) ≥7 s for men with height ≤173 cm and women with height ≤159 cm, or ≥6 s for men with height >173 cm and women with height >159 cm; (3) weak grip strength, defined as grip strength ≤29 kg for men with BMI ≤24.0 kg/m2, ≤30 kg for BMI 24.1–26.0 kg/m2, ≤30 kg for BMI 26.1–28.0 kg/m2, and ≤32 kg for BMI >28.0 kg/m2; and ≤17 kg for women with BMI ≤23.0 kg/m2, ≤17.3 kg for BMI 23.1–26.0 kg/m2, ≤18 kg for BMI 26.1–29.0 kg/m2, and ≤21 kg for BMI >29.0 kg/m2; (4) self-reported exhaustion, defined as answering “2” or “3” to either of two CES-D items (“I felt that everything I did was an effort” and “I could not get going”); and (5) low physical activity, defined as <383 kcal/week for men and <270 kcal/week for women, estimated using the Minnesota Leisure Time Activity Questionnaire (approximately equivalent to <2.5 h/week of walking for men and <2 h/week for women). According to the original definition, patients meeting ≥3 criteria were classified as frail, those meeting 1–2 criteria as prefrail/intermediate, and those meeting 0 criteria as robust. Because only 3 patients (1.1%) were robust in our cohort, robust and prefrail patients were combined into a non-frail group.
Thus, non-frail was defined as meeting 0–2 criteria (n = 173, 62.0%), and frail was defined as meeting ≥3 criteria (n = 106, 38.0%). The Fried frailty phenotype has been widely used in Chinese older adults. A modified version of this phenotype (CMFP) has been validated in Chinese older adults, demonstrating good agreement with the original phenotype (Spearman’s r = 0.795, kappa = 0.663) and excellent predictive ability for adverse outcomes (AUC = 0.936).27
Data Collection
All data were collected by two professionally trained researchers. Height and body weight were measured with patients in light clothing and barefoot, using a stadiometer and an electronic scale (Deli LE119), respectively. Frailty-related assessments were performed through face-to-face interviews and standardized physical tests.
The assessment included a 4.57-meter walking speed test, which was conducted twice, with the average value adopted for analysis. Dominant hand grip strength was measured three times using a dynamometer (Yicheng JX180), with patients seated and the elbow flexed at 90 degrees; the maximum value was included in the final analysis. Demographic and clinical information was extracted from electronic medical records.
The number of family visits was collected based on ward visit logs and verified with inpatient nurses. For ambiguous patient responses or missing data, we supplemented and verified relevant information by reviewing medical records and consulting nursing staff. The two researchers independently entered all data and performed mutual cross-checks to ensure accuracy.
Finally, 279 patients with complete data were included in the current analysis.
Statistical Analysis
Statistical analyses were performed with SPSS 25.0 and R 4.2.1. Univariate analyses, including between-group comparisons and Spearman’s rank correlation, were conducted in SPSS. For continuous variables, normality was assessed using the Shapiro–Wilk test; normally distributed variables were compared using independent-samples t-tests, while skewed variables were compared using Mann–Whitney U-tests.
Categorical variables were compared using chi-square tests. For 2 × 2 tables, Fisher’s exact test was used whenever any cell had an expected count below 5. For larger tables, it was used when more than 20% of cells had expected counts below 5 or when any cell had an expected count below 1.
We also checked for multicollinearity using the variance inflation factor (VIF).
The sample size was estimated based on the rule of thumb of at least 10 events per variable (EPV), assuming a maximum of 11 statistical variables, which required at least 110 events. All subsequent modeling and evaluation were performed in R. Univariate analysis was used as a preliminary dimensionality reduction step to avoid overfitting in the setting of a limited sample.
Variables with P < 0.1 in univariate analysis (a threshold chosen to avoid excluding potentially relevant variables) or those judged clinically important and pre-specified during the study design phase were entered into LASSO regression with 10-fold cross-validation. The most parsimonious set of variables was defined as the set of variables with non-zero coefficients selected using the lambda.1se criterion. We put these variables into a binary logistic regression (enter method) to confirm independent associated factors of frailty.
We built a nomogram in R from the final model to visualize individual frailty probability. Model performance was evaluated using ROC curves, calibration curves, and decision curve analysis (DCA). The AUC with 95% CI was calculated via DeLong’s method.
Calibration was assessed using the Hosmer–Lemeshow test and the Brier score, which was calculated on the complete dataset during bootstrap internal validation. DCA was used to evaluate clinical net benefit. Internal validation was performed using bootstrap resampling (1,000 iterations).
A two-sided P < 0.05 was considered statistically significant.
Results
Patient Characteristics and Prevalence of Frailty
We enrolled 279 older inpatients with schizophrenia, of whom 199 (71.3%) were men and 80 (28.7%) were women. According to the Fried frailty phenotype, 3 patients (1.1%) were robust (0 criteria), 170 (60.9%) were prefrail (1–2 criteria), and 106 (38.0%) were frail (≥3 criteria). Given the very small robust group (n = 3), robust and prefrail patients were combined into a non-frail group (0–2 criteria; n = 173, 62.0%), leaving 106 patients (38.0%) in the frail group (≥3 criteria).
When we compared the frail and non-frail groups, several differences emerged. Compared with the non-frail group, frail patients were older (median 69.00 vs 66.00 years, P < 0.001), had lower BMI (22.40 ± 3.55 vs 23.28 ± 3.46 kg/m2, P = 0.043), and scored worse on ADL (median 95.00 vs 100.00, P < 0.001). Frail patients also had higher rates of fall history (26.4% vs 6.4%), walking aid (15.1% vs 1.7%), somatic pain (25.5% vs 13.3%), and sensory impairment (single: 20.8% vs 13.9%; dual: 32.1% vs 8.1%) (all P < 0.05; Table 1).
No significant between-group differences were observed for the remaining variables (all P > 0.05; Table 1).Table 1Clinical Characteristics of Frail and Non-Frail Older Patients with SchizophreniaVariableNon-frail (n = 173)Frail (n = 106)χ2/t/zPSex, n (%)0.027a0.869Male124 (71.7)75 (70.8)Female49 (28.3)31 (29.2)Education, n (%)3.128a0.372Primary school or below27 (15.6)14 (13.2)Middle school78 (45.1)44 (41.5)High school or technical secondary school56 (32.4)44 (41.5)College degree or above12 (6.9)4 (3.8)Number of diagnoses, n (%)0.932a0.334<586 (49.7)59 (55.7)≥587 (50.3)47 (44.3)Number of family visits, n (%)5.153a0.076≥ 12 times/year56 (32.4)32 (30.2)6–12 times/year76 (43.9)36 (34.0)<6 times/year41 (23.7)38 (35.8)Smoking, n (%)1.009a0.315No121 (69.9)68 (64.2)Yes52 (30.1)38 (35.8)Family psychiatric history, n (%)–b1.000No169 (97.7)103 (97.2)Yes4 (2.3)3 (2.8)Number of medications, n (%)0.201a0.654<529 (16.8)20 (18.9)≥ 5144 (83.2)86 (81.1)Walking aid, n (%)18.487a< 0.001No170 (98.3)90 (84.9)Yes3 (1.7)16 (15.1)Somatic pain, n (%)6.626a0.010No150 (86.7)79 (74.5)Yes23 (13.3)27 (25.5)Surgical history, n (%)0.160a0.689No131 (75.7)78 (73.6)Yes42 (24.3)28 (26.4)Sensory impairment, n (%)33.305a< 0.001No135 (78.0)50 (47.2)Single24 (13.9)22 (20.8)Dual14 (8.1)34 (32.1)Swallowing–b0.291No170 (98.3)106 (100.0)Yes3 (1.7)0 (0.0)Fall history, n (%)21.989a< 0.001No162 (93.6)78 (73.6)Yes11 (6.4)28 (26.4)BMI, kg/m223.28 ± 3.4622.40 ± 3.552.0300.043Age, years66.00 (62.00, 70.00)69.00 (64.75, 74.00)−4.357< 0.001Disease duration, years38.00 (31.00, 44.00)40.00 (32.00, 47.25)−1.8930.058Length of current hospitalization, years7.00 (5.00, 10.00)7.00 (5.00, 20.25)−1.8440.065ADL score100.00 (95.00, 100.00)95.00 (85.00, 100.00)−5.907< 0.001Notes: Non-frail: 0–2 Fried criteria (robust + prefrail; n = 173). Frail: ≥3 Fried criteria (n = 106). Data are presented as mean ± SD for normally distributed variables, median (IQR) for skewed variables, or n (%) for categorical variables. t-test for normally distributed continuous variables; Mann–Whitney U-test (z) for skewed variables.Abbreviations: a = Chi-square test; b = Fisher exact test.
IQR = interquartile range; ADL = activities of daily living; BMI = body mass index; SD = standard deviation.
Correlation Analysis
Spearman’s rank correlation analysis showed that frailty status was significantly positively correlated with sensory impairment (rs = 0.345, P < 0.01), fall history (rs = 0.281, P < 0.01), age (rs = 0.278, P < 0.01), walking aid (rs = 0.257, P < 0.01), length of current hospitalization (rs = 0.194, P < 0.01), and somatic pain (rs = 0.154, P < 0.01). Frailty status was significantly negatively correlated with ADL score (rs = −0.293, P < 0.01) and BMI (rs = −0.121, P < 0.05).
Multivariable Logistic Regression Analysis of Factors Associated with Frailty
We used LASSO regression with 10-fold cross-validation to shrink the candidate variables, selecting the penalty parameter lambda.1se to obtain the most parsimonious model. The LASSO coefficient profiles and the cross-validation curve are shown in Supplementary Figures S1 and S2. Five clinical factors (represented by six statistical variables) were retained and entered into a binary logistic regression (enter method).
We checked for multicollinearity using VIF. All VIF values were below 2, indicating no significant collinearity among the associated factors. The final model retained five clinical factors: age, fall history, walking aid, sensory impairment, and ADL score (Table 2).
Dual sensory impairment was significantly associated with frailty (OR = 3.470, P = 0.002). ADL score showed a significant inverse association (OR = 0.967, P = 0.040). Walking aid showed a borderline significant association (OR = 4.021, P = 0.051).
Full coefficients and 95% CIs are presented in Table 2. The forest plot (Supplementary Figure S3) shows the odds ratios and 95% CIs for each factor.Table 2Multivariable Logistic Regression Analysis of Factors Associated with Frailty in Older Patients with SchizophreniaVariableBSEWald χ2POR95% CIAge0.0670.0257.1740.0071.0701.018–1.124Fall history (Yes)0.9080.4504.0790.0432.4801.027–5.987Walking aid (Yes)1.3920.7123.8210.0514.0210.996–16.228Sensory: Single0.7630.3644.3990.0362.1451.051–4.377Sensory: Dual1.2440.4079.3320.0023.4701.562–7.708ADL−0.0340.0164.2090.0400.9670.936–0.999Constant−2.4812.5050.9810.3220.084Notes: Sensory impairment was entered as a categorical variable with two dummy indicators (single vs none, dual vs none).Abbreviations: B = regression coefficient; SE = standard error; OR = odds ratio; CI = confidence interval. Variable assignments in the final model: Fall history (0 = no, 1 = yes); Walking aid (0 = no, 1 = yes); Sensory impairment (0 = none, 1 = single, 2 = dual); Age and ADL were entered as continuous variables.
ADL = activities of daily living.
Nomogram Development
The nomogram was constructed based on the final multivariable model, incorporating age, fall history, walking aid, sensory impairment, and ADL score (Figure 1). It was designed to estimate individual frailty probability using routinely available clinical data. Figure 1Nomogram for frailty identification.
The nomogram was developed from the final multivariable logistic regression model in 279 older patients with schizophrenia. It incorporates five clinical factors: age, fall history, walking aid, sensory impairment (single or dual), and activities of daily living (ADL) score. Each factor is assigned a point value on the top scale; the total points correspond to the predicted probability of frailty on the bottom scale.A nomogram made of stacked horizontal scales used to read a frailty probability from a total point score.
Scale 1 label, Points, shows tick labels 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, with no unit shown. Scale 2 label, Age, shows tick labels 50, 60, 65, 70, 75, 80, 85, 90, with no unit shown. Scale 3 label, Fall history, has categories No and Yes.
Scale 4 label, Walking aid, has categories No and Yes. Scale 5 label, Sensory impairment, has categories None, Single, Dual. Scale 6 label, Activities of daily living, shows tick labels 100, 80, 60, 40, 20, 0, with no unit shown.
Scale 7 label, Total Points, runs 0 to 220 with tick labels every 20, with no unit shown. Final scale label, Risk of Frailty, shows tick labels 0.3, 0.5, 0.7, 0.9, with no unit shown.A nomogram mapping age, fall history, walking aid, sensory impairment and activities of daily living to frailty.
Model Performance and Internal Validation
The nomogram demonstrated acceptable discrimination, with an AUC of 0.767 (95% CI: 0.708–0.826) (Figure 2). As shown in Figure 3, the calibration curve was close to the 45-degree line, indicating good agreement between predicted probabilities and observed outcomes. The Hosmer–Lemeshow test was non-significant (χ2 = 6.861, df = 8, P = 0.552).
The Brier score (0.182), calculated on the complete dataset during bootstrap internal validation, and the calibration slope (0.967) indicated a well-fitting model. Decision curve analysis showed a positive net benefit for threshold probabilities between 0.1 and 0.4 (Figure 4), indicating clinical utility within this range. Internal validation was performed using 1000 bootstrap resamples.
The optimism-corrected AUC was 0.754, suggesting acceptable internal reproducibility. External validation is warranted before routine clinical use. Figure 2Receiver operating characteristic curve.
The receiver operating characteristic curve shows the discriminative performance of the nomogram for frailty identification in older patients with schizophrenia. The area under the curve (AUC) was 0.767 (95% CI: 0.708–0.826) by DeLong’s method.A line graph showing a receiver operating characteristic curve for a nomogram frailty identification model.Figure 3Calibration curve. The calibration curve shows the agreement between predicted and observed frailty probabilities in older patients with schizophrenia.
The dashed line represents the ideal calibration. The curve was based on 1000 bootstrap repetitions (n = 279). The mean absolute error was 0.019.
The Hosmer–Lemeshow test was non-significant (P = 0.552), and the Brier score was 0.182.A line graph showing a calibration curve comparing predicted and observed probability.Figure 4Decision curve analysis. The decision curve analysis shows the net benefit of the nomogram across a range of threshold probabilities in older patients with schizophrenia. The solid blue line represents the nomogram, compared with the treat-all and treat-none strategies.
The nomogram showed positive net benefit at threshold probabilities between 0.1 and 0.4.A line graph showing net benefit versus threshold probability for model, treat all and treat none.
Discussion
This study developed a nomogram for frailty identification in older inpatients with schizophrenia. The nomogram used five routinely collected clinical factors. All variables were available from the medical record.
The tool required no specialized equipment, making it suitable for routine screening in psychiatric wards. We found a frailty prevalence of 38.0%. This was substantially higher than the 7.8% reported in Chinese community-dwelling older adults,28 but comparable to rates in general older inpatients (32.3%) and schizophrenia cohorts (32.8%).29,30 These findings align with previous observations.
Over 90% of chronic schizophrenia patients under 65 are frail or pre-frail,31 and sarcopenia-related phenotypes affect 43.6% of patients with severe mental illness.32 This discrepancy suggests that schizophrenia itself may contribute to frailty, beyond chronological aging. Although shared genetic and metabolic pathways have been proposed,20,33 these mechanisms were not directly examined in this study. Notably, even younger patients with schizophrenia can be as frail as individuals decades older.34
Sensory impairment was significantly associated with frailty in our model (dual: OR = 3.470; single: OR = 2.145). A dose-response relationship was observed: the greater the number of affected sensory domains, the higher the risk of frailty. The UK Biobank study reported that dual sensory impairment was associated with a more than twofold increased risk of frailty (OR = 2.22).35 A pooled analysis of four national cohorts further demonstrated that dual sensory impairment was associated with a 1.5 to 2.8-fold increased risk of cognitive frailty.36 Vision and hearing loss affect balance and gait directly, while reduced environmental input may accelerate cognitive decline.36,37 Fear of falling may also lead to activity avoidance, creating a cycle of disuse and further decline.38 In psychiatric settings, sensory deficits are rarely self-reported and easily overlooked.
These findings suggest they may aid risk stratification, but routine screening requires cautious interpretation and further study.
Sensory loss is not the only marker of physical decline. Fall history (OR = 2.480) and walking aid (OR = 4.021) were additional markers. Walking aid showed borderline significance (P = 0.051), suggesting a possible association that warrants further investigation.
Slow gait has repeatedly been linked to falls in schizophrenia.30 Among frailty components, slow gait appears to be a major driver of fall risk in hospitalized patients.12 Gait assessment may therefore be useful. Sarcopenia and frailty independently predict falls in stable schizophrenia patients (OR = 3.24 and 1.46, respectively),39 reinforcing the value of functional assessment in fall prevention.40 Falls and frailty have a bidirectional relationship. Reduced activity after a fall accelerates deconditioning, which in turn increases the risk of further falls.38 These findings underscore the importance of comprehensive functional assessment in this population.
Higher ADL scores were associated with lower frailty risk (OR = 0.967). However, this association should be interpreted cautiously. The median length of current hospitalization in our sample was 7 years, far longer than typical inpatient populations.
Prolonged hospitalization in a closed, low-stimulus environment may itself accelerate functional decline among these older inpatients. Ishibashi et al reported that over 60% of long-term psychiatric inpatients (mean stay 10.5 years) had severe locomotive syndrome.41 Hospital-associated deconditioning affects approximately 37% of older inpatients,42 and social function deficits affect over 70% of long-term stable schizophrenia inpatients.43 These findings suggest that current care models emphasizing “substitutive care” may inadvertently deprive patients of opportunities to maintain functional independence. A shift toward “enabling care” may help preserve functional reserve.44
Chronological age was independently associated with frailty (OR = 1.070). Frailty has been reported to progress faster in schizophrenia,6 and epigenetic clocks indicate accelerated aging across multiple systems.45 Mental and behavioural disorders are associated with biological age acceleration, and the association strengthens with each additional comorbidity.46 Frailty indices may reflect this cumulative physiological burden. These findings suggest that chronological age alone may not adequately capture frailty risk in this population.
Biological age and functional status should also be considered.
Several more comprehensive models have been developed. Lyu et al used 10 comprehensive geriatric assessment (CGA) items and achieved AUCs above 0.89.47 The Groningen Frailty Indicator (GFI) and the Tilburg Frailty Indicator (TFI), each with 15 items, report AUCs around 0.82 and 0.79.48 These tools outperform our model in discrimination. However, they require longer assessment time and additional training.
In contrast, all five associated factors in our nomogram can be extracted from the electronic medical record (EMR). This enables automatic risk calculation in the future, making our model particularly suitable for routine frailty screening in busy psychiatric wards where assessment time is limited.
This study has several limitations. First, the cross-sectional design precludes causal inferences. Second, the single-center design, convenience sampling, and predominantly male sample (71.3%) may have introduced selection bias and limit generalizability.
Although women generally have a higher risk of frailty,49 our sample included too few women to permit reliable sex-stratified analysis. Third, using a dual criterion (statistical threshold plus clinical judgment) for variable inclusion may have introduced some subjectivity. Fourth, the multistep variable selection process may produce optimistic estimates.
Fifth, some associated factors (eg, ADL, walking aid, and fall history) are conceptually related to components of the Fried frailty phenotype, which may introduce circularity or information leakage. Bootstrap internal validation showed minimal optimism (corrected AUC 0.754 vs original 0.767), but this supports internal reproducibility only and does not demonstrate generalizability. Finally, we lacked information on several potentially important variables, including antipsychotic dosing, anticholinergic burden, cognitive function, and biomarkers.
These have been linked to frailty or its progression in other studies,7–9,16,17 but we could not assess their contribution relative to our five clinical factors. External validation in independent multicenter cohorts is still lacking.
These findings have several clinical implications. Among the statistically significant factors, dual sensory impairment showed a strong association with frailty. This suggests that routine vision and hearing assessment may be useful in geriatric psychiatric evaluation.35 Patients with a fall history may benefit from gait assessment, medication review, and environmental safety checks.12,15 Higher ADL scores were associated with lower frailty risk, supporting further evaluation of physical activity in care planning.23,24,50 However, external validation is warranted before routine clinical application of the nomogram.
Future prospective, multicenter studies are needed to validate its clinical utility and generalizability.
Conclusions
The prevalence of frailty in this study was 38.0%. The nomogram incorporated five routinely available clinical factors: age, fall history, walking aid, sensory impairment, and ADL score. The model showed acceptable discrimination and good calibration.
All five clinical factors can be extracted from the electronic medical record (EMR). Therefore, the nomogram may serve as a practical tool for frailty identification in psychiatric wards, pending external validation. Because this was a cross-sectional study, the findings support identification of current frailty rather than prediction of future risk or causal inference.
Prospective multicenter external validation is warranted before routine clinical use.