Background: Falls are the leading cause of fatal and non-fatal injuries in older adults, and attention to falls prevention is imperative. Prognostic models identifying high-risk individuals could guide fall-preventive interventions in the rapidly growing older population. We aimed to develop a prognostic prediction model on falls rate in community-dwelling older adults.
Methods: Design: prospective cohort study with 12 months follow-up and participants recruited from June 14, 2018, to July 18, 2019.
Setting: general population.
Subjects: community-dwelling older adults aged 75+ years, without dementia or acute illness, and able to stand unsupported for one minute.
Outcome: fall rate for 12 months.
Statistical methods: candidate predictors were physical and cognitive tests along with self-report questionnaires. We developed a Poisson model using least absolute shrinkage and selection operator penalization, leave-one-out cross-validation, and bootstrap resampling with 1000 iterations.
Results: Sample size at study start and end was 241 and 198 (82%), respectively. The number of fallers was 87 (36%), and the fall rate was 0.94 falls per person-year. Predictors included in the final model were educational level, dizziness, alcohol consumption, prior falls, self-perceived falls risk, disability, and depressive symptoms. Mean absolute error (95% CI) was 0.88 falls (0.71-1.16).
Conclusion: We developed a falls prediction model for community-dwelling older adults in a general population setting. The model was developed by selecting predictors from among physical and cognitive tests along with self-report questionnaires. The final model included only the questionnaire-based predictors, and its predictions had an average imprecision of less than one fall, thereby making it appropriate for clinical practice. Future external validation is needed.
Trial registration: Clinicaltrials.gov ( NCT03608709 ).
Keywords: Accidental falls; Models; Multivariable analysis; Prognosis; Theoretical.