Predicting the risk of post-discharge medication related harm in older adults

Student thesis: Doctoral ThesisDoctor of Philosophy


Older people are at an increased risk of experiencing medication related harm (MRH), but it is not clear which older people are at greatest risk. Multivariable risk prediction models target interventions at patients stratified by their risk of developing a specific outcome over time. An MRH risk prediction model could prioritise care for high-risk older patients at the point of discharge - a time of heightened patient vulnerability. The overall aim of this thesis was to examine the possibility of identifying older patients at risk of MRH following discharge from hospital. To achieve this, three distinct studies were conducted; Study 1 was a systematic review of MRH risk prediction models which informed the development of a conceptual framework and the design of Study 2. Study 2 involved conducting a prospective cohort study to determine the incidence, and describe the characteristics, of MRH in an older population discharged from a United Kingdom (UK) teaching hospital. Patients discharged from acute care were recruited and followed up for 8 weeks. A review of any re-admission, General Practice (GP) notes, and a patient telephone interview were used to determine if MRH had occurred. Likelihood, preventability, severity and the medicine(s) involved were determined. MRH was defined as harm from medicines due to an adverse drug reaction or failure to receive a medicine (including non-adherence). Study 3 identified variables associated with MRH focussing on physiological and psychosocial domains. Following an extensive literature search of over 13,000 articles, Study 1 identified only five risk prediction models of modest quality, all focussing on adverse drug reactions (ADRs), mainly drawing on physiological variables to predict medication harm and none related to the post-discharge period. Four had undertaken some validation but displayed only modest performance (Area Under Receiver Operator Curve (AUROC) 0.62-0.73), rendering them unsuitable for routine clinical use. This systematic review and a review of the wider literature around MRH led to the design of a conceptual framework, drawing together physiological and psychosocial variables and highlighting commonalities between MRH and frailty to build a theoretical model of the factors involved in medication harm. The prospective cohort study recruited 396 patients during their in-patient stay in the older persons’ unit, (mean age 83±7 years, Charlson comorbidity index (CCI) 2.16±1.63, mean Barthel index 13.24±4.56, 55% received a package of care) and highlighted that two in five frail patients experienced MRH, 88% of which required treatment modification and more than half were preventable. Approximately a third required re-admission. Non-adherence contributed to approximately one third of the MRH incidents. Medicines acting on the central nervous system were responsible for the majority of MRH (25%). Physiological and psychosocial variables were associated with MRH at a univariate level: number of medicines (odds ratio (OR) 1.10 [95% confidence interval (CI) 1.05-1.16, p<0.0001]), antithrombotics (OR 1.71 [95%CI 1.12-2.60, p0.01]), antihypertensives (OR 1.55 [95%CI 1.03-2.33, p0.04]), MUST score (OR 1.92 [95%CI 1.12-3.29, p0.02]), and hand grip strength (men) (OR 0.93 [95%CI 0.89-0.98]). A higher than expected re-admission rate was reported; 156 patients (39%) were re-admitted, 67 with MRH. Sub-analysis revealed a mean time to event of 48.9 days (95% CI 47.3-50.6) and when compared to those not re-admitted, identified additional variables associated with MRH: use of a multi-compartment compliance aid (MCA) (chi square 9.76, p 0.002), CCI (hazard ratio (HR) 1.14 [95% CI 1.01-1.29, p0.03]) and depression (HR 1.04 [95% CI 1.01-1.08, p0.02]). Number of medicines was also significant (HR 1.10 [95% CI 1.05-1.15, p<0.01]) and was the only variable to retain significance upon multivariate analysis. There was little difference between patients re-admitted with MRH and patients re-admitted without MRH with only number of medicines reaching statistical significance (OR 1.08 [95% CI 1.01-1.17, p0.04]). All re-admitted patients were frail. The heterogeneity of an older population, due to advancing age and frailty, makes the identification of a small number of highly predictive variables challenging and may limit the usefulness of pursing the traditional path of risk prediction modelling. The study population were frail and given the high prevalence of low reserve, the significance of the number of medicines may represent both a greater illness burden, and increased likelihood of exposure to energies may be better expended exploring the clinical utility of a frailty index to stratify the risk of MRH in this complex population. potential medicine related insult thus an increased risk of MRH. This study suggests that MRH is similar to geriatric syndromes; it is multifactorial, occurs when there is an accumulation of deficits across multiple systems, and renders the patient vulnerable to situational challenges. In light of this,
Date of Award2017
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorFinbarr Martin (Supervisor) & Graham Davies (Supervisor)

Cite this