Abstract
BACKGROUND: Inactive behaviour is common in older adults during hospitalisation and associated with poor health outcomes. If patients at high risk of spending little time standing/walking could be identified early after admission, they could be given interventions aimed at increasing their time spent standing/walking. This study aims to identify older adults at high risk of low physical activity (PA) levels during hospitalisation. METHODS: Prospective cohort study of 165 older adults (≥ 70 years) admitted to the department of Internal Medicine of Maastricht University Medical Centre for acute medical illness. Two prediction models were developed to predict the probability of low PA levels during hospitalisation. Time spent standing/walking per day was measured with an accelerometer until discharge (≤ 12 days). The average time standing/walking per day between inclusion and discharge was dichotomized into low/high PA levels by dividing the cohort at the median (50.0%) in model 1, and lowest tertile (33.3%) in model 2. Potential predictors-Short Physical Performance Battery (SPPB), Activity Measure for Post-Acute Care (AM-PAC), age, sex, walking aid use, and disabilities in activities of daily living-were selected based on literature and analysed using logistic regression analysis. Models were internally validated using bootstrapping. Model performance was quantified using measures of discrimination (area under the receiver operating characteristic curve (AUC)) and calibration (Hosmer and Lemeshow (H-L) goodness-of-fit test and calibration plots). RESULTS: Model 1 predicts a probability of spending ≤ 64.4 min standing/walking and holds the predictors SPPB, AM-PAC and sex. Model 2 predicts a probability of spending ≤ 47.2 min standing/walking and holds the predictors SPPB, AM-PAC, age and walking aid use. AUCs of models 1 and 2 were .80 (95% confidence interval (CI) = .73-.87) and .86 (95%CI = .79-.92), respectively, indicating good discriminative ability. Both models demonstrate near perfect calibration of the predicted probabilities and good overall performance, with model 2 performing slightly better. CONCLUSIONS: The developed and internally validated prediction models may enable clinicians to identify older adults at high risk of low PA levels during hospitalisation. External validation and determining the clinical impact are needed before applying the models in clinical practise.