The research was conducted as a prospective, single-center cohort study between December 2024 and February 2025 at the Cardiovascular Disease Center of a tertiary-care hospital in China. The study was conducted and reported following the TRIPOD statement, which guides transparent reporting of multivariable prediction models for prognosis or diagnosis (13).

A convenience sampling method was employed to enroll older patients with acute HF. The standards for including participants in this research were as follows: (1) HF as the primary reason for hospitalization; (2) admission to the general wards; (3) age ≥ 60 years; and (4) provision of informed consent. The exclusion standards were as follows: (1) discharge from the hospital within 48 h; (2) complete dependence in ADLs, which precludes the further functional decline required for developing HAD; (3) significant hearing, visual, or communication impairments; and (4) a life expectancy of less than one month.

The minimum sample size was calculated using the framework by Riley et al. for multivariable prediction models with binary outcomes (14). Assumptions included a 30% HAD incidence based on a meta-analysis of older patients (15), and an area under the receiver operating characteristic curve (AUC) of 0.87 for model discrimination (16). Accordingly, the Cox-Snell R² was estimated to be 0.34 using the approach described in the methodological literature (17). Furthermore, a total of 12 predictors were assumed, including 8 variables reported in a study of general medical inpatients (18) and 4 additional variables intended to capture HF-specific characteristics and potential psychosocial indicators during hospitalization. This resulted in a minimum sample size of 323, deemed sufficient to ensure robust model development with adequate calibration and generalizability.

HAD was defined as a reduction of 5 points or more in a patient's capacity to carry out ADLs, as measured by the Barthel Index (BI) at admission and discharge intervals. The BI, a well-established 10-item measurement scale, evaluates functional independence across two critical dimensions: self-care and mobility competencies (19). The BI categorizes patients into the following clinical dependency levels: complete independence, which is scored as 100; mild dependence, corresponding to scores ranging from 61 to 99; moderate dependence, with scores falling between 41 and 60; and severe dependence, indicated by scores of 40 or less.

Data analysis was conducted within the R statistical environment (version 4.4.2) (28). Since missingness was limited to covariates and affected merely 6 patients (1.23%), a rate below 5%, a complete-case analysis was appropriately employed (29, 30). Descriptive statistics summarized categorical and continuous variables. Categorical data were analyzed using frequency distributions and chi-square tests. Continuous data were assessed for normality using the Shapiro–Wilk test, with P-values below 0.05 indicating non-normality. Since all continuous variables were non-normally distributed, medians (IQRs) were reported, and between-group differences were evaluated with the Mann–Whitney U-test.

Participants were randomly assigned to training and testing sets in a 7:3 ratio using simple random sampling (31). A multi-stage variable screening strategy was employed to select strong predictors for nomogram construction, with HAD serving as the dependent variable throughout the entire process. The process began with univariable logistic regression to assess the association between each candidate variable and HAD, retaining variables with P < 0.05 for further analysis. To address potential multicollinearity among continuous variables, Spearman's rank correlation was used (32). Any variable pairs with a correlation coefficient exceeding the pre-defined threshold of 0.8 would be reviewed, with the variable deemed to have lesser clinical relevance or weaker univariable statistical importance being removed. Remaining variables underwent Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection, with the optimal λ determined via 10-fold cross-validation (33). Variables with non-zero coefficients in the LASSO model were then included in a multivariable logistic regression model, where associations were quantified as odds ratios (ORs) with 95% confidence intervals (CIs). The final predictors for the nomogram were those with statistically significant associations in the multivariable model.

The nomogram's performance was assessed in terms of discrimination, calibration, and clinical utility. Discrimination was evaluated using receiver operating characteristic curve analysis, with the AUC as the primary indicator; an AUC > 0.75 indicated satisfactory discrimination (34). Calibration was evaluated using three approaches. Calibration curves visually assessed the agreement between predicted probabilities and observed outcomes. The Hosmer-Lemeshow test assessed goodness-of-fit, with a non-significant result indicating adequate fit. The Brier score quantified overall prediction accuracy, with a score below 0.25 indicating good calibration (35). Clinical utility was evaluated through decision curve analysis (DCA), which quantified net benefit across various threshold probabilities. Additionally, a clinical impact curve (CIC) was constructed to illustrate the number of high-risk individuals classified and true positives identified at each threshold.

Approval for this research protocol was granted by the Ethics Committee of the School of Nursing, Jilin University (No. 2024120301). Prior to enrollment, written informed consent was obtained from all participants.

Table 1 summarizes the demographic, clinical, behavioral, and functional characteristics of the 480 participants. The cohort had a median age of 75.50 years, with females comprising 56.88%. Most patients (72.92%) had HF secondary to ischemic or hypertensive heart disease, and the median length of hospital stay was 9 days. By discharge, 201 participants developed HAD, yielding an incidence of 41.88%. Participants were randomly assigned to training (n = 336) and testing (n = 144) cohorts, which differed significantly in educational attainment (P = 0.040), living arrangements (P = 0.031), and rating of mobility (P = 0.047). Comparison between HAD and non-HAD groups revealed significant differences in demographic, clinical and functional characteristics, such as age, EF, route of admission, and scores on the IADLs scale (see Appendix Supplementary Table S1).

The 12 variables filtered out through LASSO regression method were integrated into a multivariate logistic regression analysis for further modeling. Risk factors indicating increased odds of HAD included age 81 years or older (OR = 2.34, 95 % CI = 1.175–4.663, P = 0.016), emergency admission (OR = 2.01, 95 % CI = 1.146–3.535, P = 0.015), EF of 40% or less (OR = 2.78, 95 % CI = 1.390–5.549, P = 0.004), high comorbidity burden (OR = 2.11, 95 % CI = 1.057–4.213, P = 0.034), mild cognitive impairment (OR = 2.05, 95 % CI = 1.017–4.115, P = 0.045), moderate to severe cognitive impairment (OR = 2.26, 95 % CI = 1.132–4.507, P = 0.021), high nutritional risk (OR = 2.20, 95 % CI = 1.119–4.336, P = 0.022), severe pre-admission functional dependence (OR = 2.30, 95 % CI = 1.153–4.568, P = 0.018), severe insomnia (OR = 2.36, 95 % CI = 1.185–4.700, P = 0.015), and depressive symptoms (OR = 1.92, 95 % CI = 1.093–3.380, P = 0.023). Conversely, protective factors associated with significantly decreased odds of HAD were higher scores on the rating of mobility (OR = 0.81, 95 % CI = 0.732–0.891, P < 0.001), the IADLs scale (OR = 0.72, 95 % CI = 0.626–0.818, P < 0.001), and the MOS-SSS (OR = 0.94, 95 % CI = 0.907–0.969, P < 0.001). Further details are presented in Table 2.

Guided by the multivariable logistic regression findings, a predictive nomogram incorporating 12 variables was developed to estimate HAD risk among older individuals diagnosed with acute HF (Figure 2). When utilizing the nomogram, the patient's scores for each variable should first be identified, and the corresponding points are then obtained by locating the red dots on the top “Points” axis. Subsequently, these points are added together to calculate the total point. Finally, a vertical projection from the “Total points” axis to the “Pr (HAD)” axis provides an estimate of the probability of HAD. For instance, the total point for the 15th patient was 706, corresponding to an estimated 38.6% probability of developing HAD.

The nomogram demonstrated good discrimination, with an AUC of 0.841 (95% CI = 0.797–0.884) (Figure 3A). Calibration analysis indicated strong agreement between predicted and observed HAD probabilities (mean absolute error = 0.037; Figure 3C). This was further supported by a non-significant Hosmer-Lemeshow test (χ² = 7.75, P = 0.459) and a Brier score of 0.155, reinforcing the model's good calibration. Moreover, DCA revealed that across most of the threshold probability range (0–1), the nomogram provided a superior net benefit compared with two extreme clinical strategies illustrated in Figure 3E, namely the “All” strategy represented by the red line, which involves treating all individuals regardless of their risk, and the “None” strategy represented by the black line, which involves treating no individuals. The CIC graphically reinforced this utility (Figure 3G). It demonstrated that, particularly when the threshold probability exceeded 0.6, the number of individuals identified as high-risk closely matched actual HAD cases. These results support the nomogram's robust performance and clinical utility.

Internal validation through a 30% stratified random subset demonstrated the nomogram's robust predictive performance. The AUC for the testing cohort was 0.786 (95% CI = 0.710–0.862) (Figure 3B). The calibration plot illustrated a remarkable consistency between the estimated probabilities of HAD and the actual ones, quantified by a mean absolute error of 0.026 (Figure 3D). This strong calibration was confirmed by a non-significant Hosmer-Lemeshow test result (χ² = 6.85, P = 0.553) and supported by a Brier score of 0.188. Furthermore, based on the DCA, the nomogram provided a higher net benefit across a broad range of threshold probabilities, outperforming both the “All” strategy (red line) and the “None” strategy (black line) (Figure 3F). The CIC analysis corroborated this finding, showing that at high-risk thresholds above 0.6, the number of predicted cases closely aligned with the actual number of events (Figure 3H). Together, these findings support the favorable clinical utility of the nomogram for guiding HAD prediction and intervention decisions.

The incidence of HAD in older Chinese patients with acute HF was 41.88%, higher than the 37.1% reported in a recent Japanese multicenter cohort study (6). This difference can be explained by several factors. First, the median EF in our cohort was 43.00%, notably below the 48.0% reported in the Japanese study, while the median pre-admission BI was 60.00, significantly lower than their 88.3. These differences indicate that patients in this study experienced more severe cardiac dysfunction and possessed a poorer baseline physiological functional reserve, both of which significantly increased the likelihood of developing HAD, as demonstrated by the constructed nomogram. Furthermore, variations in inpatient management practices may also contribute to this disparity. Participants in the Japanese study engaged in standardized inpatient cardiac rehabilitation protocols, incorporating early low-intensity mobilization strategies, progressive endurance exercises, and structured resistance training regimens. This individualized, multicomponent strategy may have contributed to shorter bed-rest times, fostering functional recovery and thereby mitigating the risk of HAD.

Reduced EF and emergency admission, which were identified as independent predictors of HAD through multivariable logistic regression, were subsequently visualized in the nomogram to serve as indicators of disease severity. A novel contribution of the present study is the emphasis on the important role of cardiac function in the development of HAD. An EF of 40% or less denotes a marked decline in cardiac output, driving chronic hypoperfusion that accelerates skeletal muscle ischemia and loss (36). Although direct quantification of muscle strength or mass was not performed, prior studies reported a 2.5-fold higher sarcopenia risk among patients with reduced EF compared with those with preserved EF (37), implying a diminished muscular reserve in the reduced-EF subgroup. Recent evidence also indicates that HF patients with preserved EF are at a similarly high risk of developing HAD. In populations primarily characterized by this phenotype, factors such as advanced age, a high burden of comorbidities, significant frailty, and nutritional risks contribute to a heightened baseline vulnerability to functional decline (38). These findings suggest that the mechanisms leading to HAD may vary depending on the phenotype. In patients with reduced EF, the risk is predominantly driven by cardiac dysfunction and its systemic effects, whereas in those with preserved EF, the risk is more closely linked to broader vulnerabilities, such as geriatric syndromes, multimorbidity, and impaired physical capacity. Emergency admission represents an acute dimension of risk. Consistent with our findings, prior geriatric research demonstrated that emergency admission approximately doubled the odds of HAD compared with elective admission (8). This association may be explained by the sudden worsening of HF symptoms that triggers a cascade of stress responses such as neuroendocrine dysregulation and immune imbalance (39), which hinder recovery and accelerate functional decline, as described in the Nosocomial Based Stress Model (10). Furthermore, urgent therapeutic measures (e.g., diuretics for pulmonary edema) may induce electrolyte shifts such as hypokalemia, resulting in fatigue and dizziness that limit patient mobility and exacerbate in-hospital muscle deconditioning (40).

The nomogram integrated six critical variables reflecting baseline physiological functional reserve, including age, comorbidity burden, cognitive function, nutritional risk, and pre-admission functional status in ADLs and IADLs. Among these variables, cognitive impairment emerged as a particularly influential predictor, wherein even mild deficits significantly increased the risk of HAD. This finding aligns with prior studies focusing on older cardiovascular patients, which established cognitive impairment as an independent predictor of HAD (41, 42). In particular, Honda et al. reported that cognitive impairment, as a univariate predictor, achieved an AUC of 0.77, demonstrating its discriminative power and critical role in HAD development (42). Nutritional risk was also retained, consistent with earlier work that identified loss of appetite as a predictor in cardiac patients (41). A noteworthy study suggested that the CONUT score, an objective indicator derived from serum albumin, total cholesterol, and total lymphocyte count, may offer superior predictive value for HAD in older patients with acute HF compared to tools relying on self-reported items, highlighting the potential utility of biomarker-based nutritional assessments in this population (43). The remaining predictors, including advanced age, multiple comorbidities, pre-admission functional dependence in ADLs and IADLs, have all been established as statistically significant predictors of HAD in previous cohort studies (44–46). Overall, these results support the multidimensional cumulative impairment paradigm, in which co-occurring deficits across physiological systems synergistically compromise adaptive capacity and heighten vulnerability to functional decline following acute illness or iatrogenic stressors (3).

The nomogram further incorporated four variables that reflect functional status during hospitalization, including physical mobility, sleep disturbance, depressive symptoms, and perceived social support. Among these, physical mobility, typically reflected by low movement levels, emerges as one of the most prevalent deficits (47). In line with our findings, prior studies have consistently demonstrated that in-hospital immobility increases the risk of HAD regardless of the mobility metric employed (8, 46, 48, 49). Furthermore, this association has been quantitatively examined in patients with acute HF, where each additional day of delayed ambulation was independently associated with a 17% increase in the odds of impaired ADLs at discharge (50). According to current Chinese guidelines, an early mobilization protocol, comprising low-intensity resistance training, joint mobilization, and respiratory muscle training, should be initiated as soon as vital signs stabilize (51). However, in real-world clinical settings, traditional “bed-rest” culture, concerns about fall risk, and overprotective attitudes among family members often lead to a substantial gap between recommended and actual levels of mobility. Moreover, by integrating sleep disturbance, depressive symptoms, and perceived social support into a HAD prediction framework for older patients with acute HF, this study provides novel validation of the Nosocomial Based Stress Model in this cohort. Specifically, the effect sizes of these psychosocial stressors were comparable to those of established physiological indices within the multivariable model, highlighting their significant role in functional deterioration. According to the Nosocomial Based Stress Model, the mechanistic explanation for this finding is that inpatient psychosocial stressors interact synergistically with illness-induced physiological responses, amplifying neuroendocrine and immune dysfunction, while concurrently fostering a hypervigilant cognitive state that undermines rehabilitation engagement and accelerates functional decline (10). Therefore, routinely assessing these psychosocial stressors in clinical practice not only refines high-risk identification but also provides more actionable targets for preventive interventions.

This study advances both the theoretical understanding and clinical practice of HAD through the development and validation of a predictive nomogram. From a theoretical perspective, although our prior systematic review categorized HAD risk factors into baseline vulnerability (e.g., frailty) and in-hospital process factors (e.g., immobility), the role of psychosocial stressors during hospitalization has remained underexplored (52). Guided by the Nosocomial Based Stress Model, this research demonstrates that depressive symptoms and low perceived social support serve as independent predictors of HAD, thus validating and extending the applicability of this model in older adults with acute HF.

From a practical standpoint, the nomogram integrates three complementary risk dimensions, including baseline vulnerability, acute disease severity during hospitalization, and modifiable in-hospital stressors. This integration addresses a critical limitation of existing frailty assessments, which primarily capture pre-admission deficits, by enabling more accurate identification of high-risk patients, including those with relatively preserved baseline function. Consequently, the nomogram provides empirical support for a multidimensional risk framework of HAD and serves as a valuable complement to conventional clinical assessments, facilitating early, targeted preventive interventions tailored to specific risk profiles.

Several limitations should be acknowledged. First, the single-center design and convenience sampling approach has the potential to introduce selection bias, restricting the applicability of the study results to other healthcare settings or populations. Second, although the selection of a broad range of predictors was guided by literature review and theoretical rationale, it remains inherently subjective. Third, external validation for the developed nomogram was not implemented in the current research, which may introduce uncertainties regarding its performance. Fourth, while the nomogram integrates routinely assessed clinical variables, applying it at the bedside may pose practical challenges in fast-paced clinical environments without the support of a dedicated electronic tool. Despite these limitations, this study represents a meaningful step toward precise estimation of HAD risk in geriatric HF care. Future research should prioritize multicenter collaborations, integrate objective biomarkers, validate the nomogram in diverse ethnic and healthcare contexts, and focus on the development of integrated digital tools (e.g., clinical calculators or mobile applications) to facilitate point-of-care risk assessment and enhance clinical adoption, thereby further refining predictive accuracy, scientific validity, and practical applicability.