Discussion
This is the first study that compares two predictive scores in a prospective study. Mean BOOST score was 4.0 (IQR ± 1.0) and was similar to an earlier retrospective study in another London hospital of 324 patients [11]. The LACE mean score of 12 (± 2.0) is similar to other studies with slightly higher scores in those re-admitted compared to those who were not re-admitted [15].
The readmission rate was 33% in this study and older patients had higher readmission rates, this differs from others where rates of 8% to 22% have been reported [10,15]. However, neither of these studies explicitly examined older patients and thus, our readmission rate is not unexpected. The higher readmission rate may have impacted our data in lowering the number of predicted non-events in our power calculation (which used the literature readmission rate of 14% for our population. This may have led our study to be underpowered.
Examining the area under the receiver operating characteristic (ROC) curve was 0.667 for BOOST and 0.685 for LACE, both of which are significant predictors of readmission and the results are comparable to previous studies [9,10,15]. AUCs of less than 0.7 are deemed ‘good’ predictive capacity which reflects the overlap in predictions of those readmitted versus those who are not. The discriminatory ability of LACE has previously been reported as poor in the elderly population [16] and systematic review found that most models performed poorly (mainly in US population), but suggested that they may be useful with wider implementation needed [7]. This was similar to our findings with both BOOST and LACE having poor (but significant) discrimination for this group. Interestingly the systematic review, which included BOOST but not LACE, commented that few of the tools included overall health, illness severity and social determinants of health [7].
Relating to the multivariate analysis, the strongest predictor of readmission is previous admission in this cohort and hospital readmission continues to be a common phenomenon and perhaps not unexpected [9]. Given this, we perhaps need to examine the issue in a different manner and focus on transitional care following discharge from hospital for this cohort of patients as a way to prevent/reduce hospital readmission. There are now established hospital in the home and transitional programmes that allow early hospital discharge and prevent hospital admission are beneficial [17,18,19,20]. A systematic review of interventions to reduce early hospital readmissions concluded that interventions were complex with more recent ones less effective in their review from 2009 to 2013 [8]. Singaporean-based RCT showed some positive results but again highlighted the issues facing clinicians with patients with multiple comorbidity and complex care needs [21]. Co-ordination of care has been cited as an issue and this continues to be a problem in the UK [22]. The recurrent theme of previous hospitalisations as a strong indicator of future readmission, it may be that predictor models do not have benefit over a well performed history with emphasis on previous admissions. This would be an area worth future investigation.
It may be that patients require a ‘step-down’ or managed approach following hospital discharge and the role of a hospital in the home has the potential to improve post-discharge outcomes [19,23]. Within the local area, the @home service set up in 2014 manages 300-400 patients per month for short-term acute follow-up with positive results in terms of patient satisfaction [24], but did not demonstrate significant reduction in local emergency department attendances [25]. Given that hospital in the home services are now embedded into the healthcare system and integrated care is being established around the country, an exploration of targeted services for patients with high BOOST and/or LACE scores is required. This would determine if early identification of patients who had high scores and were referred for hospital in the home services translates to a lower readmission rate and better clinical outcomes for patients. Clearly, further research is needed on the various hospital in the home programmes as they are not standardised service.
A limitation of this study lies in the selection of a sub-group of patients that were admitted and discharged in a short time, mostly with length of stays less than two weeks, and not transferred to an inpatient ward. This study validates the use of the BOOST tool for recognising risk factors for readmission in these patients, but does not directly validate its use in those patients who are admitted to inpatient wards for greater lengths of stay. However, unlike the LACE score the BOOST tool does not select for length of stay and is designed only to flag patients with risk factors for readmission regardless of duration.
This single centre, prospective cohort study aimed to determine the efficacy of two models at predicting unplanned readmissions in those aged 75 and older and we have demonstrated that the mean BOOST and LACE scores for those readmitted was significantly higher than those not readmitted. Whilst the multivariate logistic regression model accurately predicted the highest number of cases with 74.5% of correct cases identified and the area under the curve was acceptable, sensitivity and specificity could be improved. Overall the predictive power is not optimal, these tools still hold some value in preventing readmissions. This study shows the strongest predictor of readmission is previous admission, and health literacy. It may be that we need to focus on education intervention to increase patient involvement in their care and ongoing management of their health. This approach has the ability to improve continuity of care and along with care coordination; there could be some benefit for decreased hospitalisations. The BOOST 2 tool has the potential to provide a pathway for quality improvement where interventions (such as teach back) based on identified risk factors (i.e. literacy) could help in preventing readmissions.
Predicting hospital readmission remains a complex task and any tool needs to be clinically relevant and reliably measured. Further prospective studies using these predictive tools may be useful in planning transitioned and hospital in the home programmes for those at high-risk of readmission.