Development of a Model Predicting 30-Day Readmission Using Prescription Information from the Medical Short Stay Units of One NHS Trust

Emergency readmission is defined within the NHS as an emergency admission to hospital
within 30 days of discharge. Excess readmissions are undesirable in terms of care quality
and efficiency; yet, despite financial incentives for improvement, reports of increasing
readmission rates continue. There is evidence that pharmacist intervention can prevent
medication errors, discrepancies and adverse drug events; which can each contribute to
readmission. The purpose of the work in this thesis was to develop a model based on
routinely collected prescription information to enable the pharmacy team to estimate
readmission risk in the clinical setting, thereby facilitating appropriate prioritisation of
potentially preventative intervention.

A multiple logistic regression model for estimating readmission risk using routinely recorded
prescription information among patients discharged home from the medical short stay units
of one NHS Trust was developed, and survival analysis was undertaken to characterise
readmission behaviour in relation to the predictors.

The readmission rate was 18% (220/1240). Readmission risk increased with increasing age
and polypharmacy: each additional medicine prescribed increased the odds of readmission
within 30 days by eight per cent and each additional year of age increased the odds of
readmission within 30 days by two per cent. Each additional medicine prescribed decreased
the time to readmission by seven per cent and each additional year of age decreased the
time to readmission by one per cent. Over one-third of readmissions occurred within one
week (73/200) and more than half (114/200) occurred within two weeks, supporting that
identification of those at risk and intervention to prevent readmission should be provided
promptly. The predictive model developed is suitable for application on admission and could
therefore enable clinicians to identify the patients most likely to require intervention to
prevent readmission before they are discharged home from hospital, thereby maximising
the time available to organise and/or provide the necessary support. Although the logistic
regression model improved accuracy by 36% compared to indiscriminate intervention whilst
identifying 70% of patients who would be readmitted, it had relatively weak discriminative
capability (c-statistic 0.637). It may be the case that clinical intuition is as effective for
predicting readmission and further research should be undertaken to confirm whether this is
the case.