This retrospective cohort study was conducted using an anonymized DPC database obtained from a large general public acute care hospital (550 beds: 520 beds on general wards and 30 beds on psychiatric wards) in Tokyo, Japan. The DPC data comprised patient-level demographic characteristics, ICD-10 codes, treatments, and prescribed drugs during all insurance-covered clinical encounters. The study used data from July 2016 to March 2021.

Patients who had been admitted to the study hospital from home or a long-term care facility (LTCF) and discharged during the study period were eligible for inclusion; the first hospitalization episode during the study period was designated the index hospitalization and included in the analysis. We excluded patients aged <65 years, patients discharged within 2 days of admission, patients who died during the index hospitalization, and patients with missing data in the study variables.

The study protocol was approved by the Ethics Committee of the Tokyo Metropolitan Geriatric Hospital and Institute of Gerontology (approval number R18-20). All procedures followed the ethical guidelines of the Medical and Biological Research Involving Human Subjects established by the Japanese government. Opt-out consent was used because all data were anonymized before being received by the authors.

The chi-square test was used to compare the differences in patient characteristics among the 3 HAC-OP-DPC categories. We generated cross-tabulations to examine the co-occurrences of each complication. Pearson correlation coefficients were calculated to measure the associations between each complication.

The associations between HAC-OP-DPC and LOS were examined using multivariable generalized linear regression models for gamma-distributed data with a log-link function that adjusted for all covariates. Effect sizes for 1 HAC-OP-DPC and ≥2 HAC-OP-DPC were quantified using adjusted risk ratios (ARRs) and their 95% CIs, which indicated the likelihood of having a longer LOS. Next, the associations of HAC-OP-DPC with discharge to other hospitals and discharge to LTCFs were examined using multivariable logistic regression models that adjusted for all covariates. Effect sizes for 1 HAC-OP-DPC and ≥2 HAC-OP-DPC were quantified using adjusted odds ratios (AORs) and their 95% CIs, which indicated the odds of being discharged to another hospital or LTCF. In addition to analyzing the associations between the number of HAC-OP-DPC and the 3 outcomes, we constructed models to examine the associations between individual HAC-OP-DPC components and the outcomes. For the analysis of hospital-associated functional decline, we excluded patients who were already dependent in all the BI components at admission because they could not experience any further functional decline. Similarly, for the analysis of hospital-associated incontinence, we excluded patients who were already dependent in both bladder function and bowel function at admission because they could not experience new-onset incontinence during hospitalization. We also conducted several sensitivity analyses. First, we re-ran the primary analysis for each of the 3 outcomes, excluding patients who could not experience functional decline (fully dependent on admission) or hospital-associated incontinence (incontinent on admission). Second, we recalculated the HAC-OP-DPC composite outcome excluding cases of delirium that were identified only by drug prescribing as we recognized that antipsychotic drugs might be prescribed for other indications (eg, behavioral and psychological symptoms of dementia, psychosis). All analyses were conducted using SPSS (version 28.0; IBM Corp), and the cross-tabulations were created using R (version 4.3.2; R project for Statistical Computing). P values (2-tailed) below .05 were considered statistically significant.

We first identified 26,780 candidate patients who were admitted to the subject hospital and discharged to home, another hospital, or an LTCF during the study period (Figure 1). After applying the exclusion criteria, the final sample for analysis consisted of 15,278 patients aged ≥65 years. Multimedia Appendix 2 shows the characteristics among 833 patients ≥65 years who died during hospitalization and without missing data at admission. These excluded participants were aged ≥85 years, had ≥2 CCI, were more likely to have ADL dependency and incontinence, and were more likely to have cancer, pneumonia, or heart failure.

Their mean age was 81.2 (SD 7.9) years, and women accounted for 55.4% of all patients (Table 1). There were 2877 (18.8%) patients who experienced 1 HAC-OP-DPC and 733 (4.8%) patients who experienced ≥2 HAC-OP-DPC during the index hospitalization.

The most common complication was functional decline (n=2103, 13.8%), followed by delirium (n=1345, 8.8%: recorded diagnosis of delirium (n=59); recorded prescriptions of drugs for delirium, n=1286), new incontinence (n=860, 5.6%), pressure injury (n=104, 0.7%), and fall (n=59, 0.4%). Figure 2 demonstrates the patterns of co-occurrence amongst HAC-OP-DPC. For example, among patients with functional decline, 20.6% also experienced incontinence during hospitalization, while 50.5% of patients with incontinence, 33.9% of patients with falls, and 22.5% of patients with delirium also experienced functional decline during hospitalization. While there were statistically significant correlations between most complications, the strength of these correlations was weak (maximum correlation coefficient of 0.26; Table 2). As shown in Table 1, older age, annual household income, multimorbidity, frailty, baseline functional impairment and incontinence, and living in LTCF were all significantly associated with HAC-OP-DPC. Patients with ≥2 HAC-OP-DPC had a significantly greater proportion of patients with congestive heart failure, pneumonia or acute bronchitis, metabolic diseases, and cancer as the principal diagnosis.

Table 3 demonstrates the significant and graded association observed between HAC-OP-DPC and outcomes. Adjusting for all covariates, those with 1 and ≥2 HAC-OP-DPC were significantly more likely to have a longer LOS during the index hospitalization (ARR 1.29, 95% CI 1.25-1.33, and ARR 1.97, 95% CI 1.87-2.08, respectively) compared to those with no HAC-OP-DPC. They were significantly more likely to be discharged to other hospitals (AOR 2.36, 95% CI 2.10-2.65, and AOR 6.96, 95% CI 5.81-8.35, respectively), and be discharged to LTCFs (AOR 1.35, 95% CI 1.09-1.67, and AOR 1.68, 95% CI 1.18-2.39, respectively). The analyses of individual HAC-OP-DPC showed that patients who experienced hospital-associated functional decline, incontinence, delirium, pressure injury, and falls were significantly more likely to have a longer LOS and be discharged to other hospitals than patients without these complications. However, only delirium was significantly associated in this individual complication analysis with discharge to LCTFs. A sensitivity analysis (n=11,075) that excluded 4203 patients who could neither experience hospital-associated functional decline nor incontinence found that HAC-OP-DPC still had a significant and graded association with longer LOS and discharge destination (Multimedia Appendix 3), similar to the results of the main analysis. Moreover, the associations of the number of HAC-OP-DPC using delirium identified from a recorded diagnosis alone with these outcomes were similar to the results of the main analysis (Multimedia Appendix 4).

This retrospective cohort study is the first to develop and apply a tool to assess HAC-OP from routinely collected administrative data and evaluate its predictive validity for hospital outcomes. Our analysis showed that almost one quarter of older inpatients with multiday stays of more than 2 days had a coded HAC-OP, and that having one or more HAC-OP-DPC was associated with longer LOS and discharge to other hospitals and LTCFs. The clinical implementation of the HAC-OP-DPC measure could support comparative analyses of clinical and policy interventions aimed at preventing these complications, thereby contributing to the optimization of acute care for older adults.

Almost 1 in 4 older patients had coding documentation of any HAC-OP-DPC in the present study, which was approximately half that of the incidence of any HAC-OP described in the Australian study [1]. Nonetheless, the incidences of each HAC-OP-DPC were ranked in similar order, with functional decline and delirium being most common. A large contribution to the disparity in incidence is likely to be reliance on DPC data to identify individual complications as HAC-OP-DPC, while the HAC-OP study used repeated patient and clinical record assessments by trained research assistants [1,17]. Although we made efforts to minimize the underestimation of delirium, falls, and pressure injuries by including additional codes and data sources in their identification criteria, it is very likely that the coding data underestimated the incidences of all complications. For example, there were no coded falls without fracture, and only 59 cases of direct recording of a delirium diagnosis. There are recognized gaps in clinician recognition and documentation of hospital complications as well as translation into coding [27]. Understanding and improving the accuracy and usability of HAC-OP-DPC may require correlation with clinical data and comparisons between sites. There are other reasons that our incidence estimates may have been lower than expected. The denominator in our main analysis included 3051 patients who could not experience further functional decline and 3803 patients who could not experience new-onset incontinence, which may have led to an underestimation of these HAC-OP-DPC. Our analyses also excluded patients who died during their inpatient stay, who may have had a higher rate of HAC-OP.

Nonetheless, our study found that the HAC-OP-DPC measure was associated with longer LOS and discharge destination, thereby demonstrating its predictive validity for outcomes in an acute care setting. Furthermore, our analysis found nonoverlapping risk estimates for LOS and discharge to other hospitals between patients with 1 and ≥2 HAC-OP-DPC, showing a significant exposure-outcome effect of a graded nature. These findings were consistent with the results of the original HAC-OP study [1] and suggest that prevention of further HAC-OP is important in those who have already acquired 1 complication. Our observations that HAC-OP-DPC was significantly associated with older age and baseline function are consistent with previous reports [1]. Importantly, our study is the first to demonstrate the strong associations of HAC-OP with higher CCI scores and HFRS, consistent with existing knowledge that comorbidities and frailty are risk factors for the individual complications included in the composite measure [13,16,34-37].

Our analysis also confirmed that the individual HAC-OP-DPC were associated with longer LOS and discharge to other hospitals, which was congruent with previous studies [6,14,35-39]. All HAC-OP-DPC have been individually recognized as important outcomes in older patients [1,6,14,17,36-42]. Although our study showed that there were patterns of co-occurrence and significant correlations among the individual HAC-OP-DPC, these correlations were weak. This suggests that these complications represent relatively distinct conditions and that they can be treated individually [1]. Moreover, a systematic review of composite outcomes in clinical trials proposed that studies should list results for all components of a composite outcome to avoid confusion and bias [43], and an outcome study using the original multicomponent HAC-OP measure demonstrated significant and clinically important improvements in individual outcomes but not in the composite measure [17]. We recommend that future studies should report the effects of interventions not only on the HAC-OP-DPC measure as a composite outcome but also on its individual complications.

Strengths of this study include a large, representative dataset with high levels of item completeness (including functional variables), adjustment for important covariates, and use of sensitivity analyses to explore data assumptions. We also recognize several limitations. First, we included additional information to reduce these anticipated underestimates that may have reduced precision; for example, by including drug prescribing of antipsychotics in the delirium diagnosis, we may have included some patients with other indications such as behavioral and psychological symptoms of dementia. However, our sensitivity analysis using a more stringent definition of delirium suggests that this had minimal impact on our overall findings. Second, our study was conducted in a single acute care hospital in Japan, and its findings may not be generalizable to other hospitals, regions, or countries. Nevertheless, the majority of acute care hospitals in Japan have adopted the DPC system [20], and future studies could compare the incidence of HAC-OP-DPC in each acute hospital throughout Japan. Also, although a previous study identified the validity of chronic disease diagnoses in DPC data as being generally high [26], we recognize that the performance of comorbidity scoring methods based on administrative data may vary between health systems [44]. Third, we could not infer a causal relationship between HAC-OP-DPC and LOS because we were unable to identify the date when each complication occurred, and we cannot exclude reverse causality between HAC-OP-DPC and LOS [1]. Fourth, our study did not examine the association between HAC-OP-DPC and mortality, which had been analyzed in the original HAC-OP study [1], because there are no BI scores assigned on discharge for in-hospital deaths to calculate functional decline and incontinence of HAC-OP-DPC, and DPC data lack information on death after discharge. Future studies could link DPC data with mortality data to examine the associations between HAC-OP-DPC and mortality. Finally, our database lacked information on potentially important confounders, such as residential status (eg, living alone or with others), the presence of caregivers, and disease severity that could be included in future studies.

This study showed that almost one quarter of older acute care inpatients in a Japanese hospital have coding indicating a HAC-OP and demonstrated the predictive validity of the HAC-OP-DPC measure for longer LOS and discharge to other hospitals and LTCFs. To attain a more robust understanding of these relationships, additional studies are needed to verify our findings in other hospitals and regions. The clinical implementation of HAC-OP-DPC, which are identified using routinely collected administrative data, could support the efficient evaluation of integrated interventions aimed at optimizing inpatient care for older adults.