Falls after hospital discharge among older adults : temporal pattern in risk, associated factors, and healthcare burden

Abstract

Global aging leads to a considerable proportion of older individuals requiring hospitalization. After hospital discharge, they are physically frail and susceptible to post-hospital falls. This risk is expected to decline after peaking in the early post-hospital stages, yet which period with the highest risk is uncertain. Limited evidence on factors associated with post-hospital falls is available. To effectively prevent post-hospital falls in older adults, it is crucial to thoroughly investigate the temporal pattern of risk, identify the period with the highest risk, estimate the healthcare economic burden, and identify dominant associated factors. Meanwhile, accurate and reliable fall identification methods by electronic medical records (EMRs) can facilitate interventions. Nevertheless, limited evidence exists regarding these areas.

This thesis aims to (i) systematically review the existing findings on incidence and risk factors for post-hospital falls in older adults; (ii) examine incidence rates of falls during varying posthospital periods, investigate the temporal pattern in risk, and quantify the related healthcare costs; (iii) identify the associated factors from hospitalization and subsequent care for posthospital falls; and (ⅳ) develop and validate a natural language processing (NLP) algorithm to detect falls in clinical notes and explore the optimal identification strategies based on EMRs.

First, a systematic review and meta-analysis was performed to synthesize the evidence on incidence and risk factors for post-hospital falls in older adults. Then, two territory-wide and population-based retrospective cohort studies were conducted utilizing 12-year EMRs (2007 to 2018) from older patients aged 65 or over and discharged from public hospitals in Hong Kong, which were obtained from the Hospital Authority Data Collaboration Lab. Furthermore, an NLP algorithm was developed and validated based on inpatient clinical notes of this cohort to enhance fall identification in EMRs.

The systematic review revealed a pooled incidence proportion of fallers for post-hospital falls in older adults of 14% based on 18 studies from eight countries. Twenty-six risk factors for post-hospital falls were identified, with the highest risks associated with previous falls, previous fractures, delirium, and neurological diseases from the biological domain. The first cohort study showed an annual incidence proportion of fallers of 4.7% and an incidence rate of 57.4 per 1000 person-years of post-hospital falls in Hong Kong. The incidence rate was highest in the first three weeks after discharge and gradually decreased to a stable level from the fourth to ninth week. The annual healthcare costs related to post-hospital falls exceeded HKD 176.0 million (28.9 million USD PPP) in older adults, with the mean cost per faller and fall being HKD 67,729 (11,129 USD PPP) and HKD 58,403 (9,596 USD PPP), respectively. The second cohort study showed that in the hospitalization and subsequent care domain, patients discharged from non-surgical wards, with a length of stay (LOS) over two weeks, and receiving the Geriatric Day Hospital and Rehabilitation Day Program were associated with increased post-hospital fall risk in Hong Kong. The developed NLP algorithm demonstrated excellent performance, with a sensitivity, specificity, precision, and F-measure of 93.3%, 99.0%, 87.5%, and 0.903 at the record and episode levels, and 92.9%, 98.3%, 89.7%, and 0.912 at the patient level.

The findings implied that fall prevention efforts could focus on the high-risk period of the first nine weeks after discharge to maximize the effect. Resource allocation should prioritize older adults with higher post-hospital fall risks, such as those with previous falls, previous fractures, delirium, neurological diseases, and LOS over two weeks. The NLP for fall identification would benefit post-hospital prevention initiatives and services by screening fallers for targeting and outcome surveillance. (580 words)