The PEARL score predicts 90-day readmission or death after hospitalisation for acute exacerbation of COPD

Background One in three patients hospitalised due to acute exacerbation of COPD (AECOPD) is readmitted within 90 days. No tool has been developed specifically in this population to predict readmission or death. Clinicians are unable to identify patients at particular risk, yet resources to prevent readmission are allocated based on clinical judgement. Methods In participating hospitals, consecutive admissions of patients with AECOPD were identified by screening wards and reviewing coding records. A tool to predict 90-day readmission or death without readmission was developed in two hospitals (the derivation cohort) and validated in: (a) the same hospitals at a later timeframe (internal validation cohort) and (b) four further UK hospitals (external validation cohort). Performance was compared with ADO, BODEX, CODEX, DOSE and LACE scores. Results Of 2417 patients, 936 were readmitted or died within 90 days of discharge. The five independent variables in the final model were: Previous admissions, eMRCD score, Age, Right-sided heart failure and Left-sided heart failure (PEARL). The PEARL score was consistently discriminative and accurate with a c-statistic of 0.73, 0.68 and 0.70 in the derivation, internal validation and external validation cohorts. Higher PEARL scores were associated with a shorter time to readmission. Conclusions The PEARL score is a simple tool that can effectively stratify patients' risk of 90-day readmission or death, which could help guide readmission avoidance strategies within the clinical and research setting. It is superior to other scores that have been used in this population. Trial registration number UKCRN ID 14214.


CHARMS CHECKLIST
Source of Data 1) Source of data (e.g. cohort, case-control, randomised trial participants, or registry data)  The derivation and external validation cohorts were prospective. The internal validation cohort was retrospective.
Participants 1) Participant eligibility and recruitment method (consecutive participants, location, number of centres, setting, inclusion and exclusion criteria) Eligible patients analysis  Patients who did not survive to discharge were appropriately excluded as the main outcome was readmission/ death without readmission at 90 days from discharge. Otherwise, all patients were included in the derivation cohort. The primary outcome for the three cohorts was for inpatient mortality, which led to the development of the DECAF score. The readmission analysis was a prespecified study. In the validation cohort, those that did have complete data for all DECAF indices were not included in the analysis, although this was only 1% of the population, and mainly comprised of patients that had oxygen saturations sufficiently low to warrant arterial blood gas analysis but that declined this investigation. Eligible patients excluded  Exclusion criteria were few. For the internal validation cohort, patients were not eligible as follows: survival <1 year n=27 (twelve lung cancer, three end stage dementia, three metastatic cancer, two metastatic bladder cancer, two idiopathic pulmonary fibrosis, one metastatic renal cancer, one metastatic bower cancer, one metastatic rectal cancer, one oesophageal cancer, and one mesenteric cancer patient), less than ten pack year smoking history n=24, spirometry not obstructive= 42. Ten patients had no ABG results, but had supplemental oxygen or oxygen saturations that were too low to assume a DECAF acidaemia score of zero. One patient had no eosinophil count. Robust data for the derivation and external validation cohort is unavailable. Consecutive patients  Extensive efforts were made to capture consecutive patients, including a broad coding search. Patients were captured by daily screening (Monday to Friday) on admission units and medical wards (derivation and external validation cohorts) by a dedicated team. In the internal validation cohort, patients were mainly identified retrospectively using a broad coding search, with cross referencing to clinical staff whose role it is to review patients with exacerbation of COPD. In the internal validation cohort, a dedicated team screened the admission units and medical wards for three months and compared patient capture to the coding records search and clinical team capture. Only one patient was identified by daily screening that was missed by coding or the clinical team. Location, centres, setting, and inclusion and exclusion criteria  Six UK centres were involved: the same two sites that were included in the derivation cohort took part in the internal validation cohort, and four geographical distinct hospitals took part in the external validation cohort. All patients in the study were recruited from secondary care. Inclusion and exclusion criteria are described.

2) Participant description
 Detailed description of participants by different sites in DECAF validation study. 18 A detailed description of patients in each cohort is shown in table 2.

3) Details of treatments received, if relevant
 Treatments to reduce hospital readmission include smoking cessation, inhaled corticosteroids, longacting beta agonists, and long-acting muscarinic agonists, and pulmonary rehabilitation. The score is intended to be used to inform management, so including many acute treatments as predictors is not appropriate. Long term oxygen and long term prednisolone were included in model development. NIV treatment is based on fairly objective criteria based on pH, and has been previously shown to be predictive, so was included (see table 2). The research team did not influence clinical treatment.

4) Study dates
 Dates for recruitment period of each hospital discussed in previous publications. 17 18 Outcome to be predicted 1) Definition and method for measurement of outcome  Readmission/ death without readmission 90 days from discharge in patients surviving to discharge. Readmission was clearly defined-a patient had to be admitted to hospital and reviewed by a member of the clinical team.
2) Was the same outcome definition (and method for measurement) in all patients?  Yes.
3) Type of outcome single or combined endpoints?
 Combined outcome. We did not wish to create a score that identified those at risk of readmission, but missed those at risk of death without readmission, as some of these deaths may be preventable. Predictors of readmission and death are similar. 4) Was the outcome assessed without knowledge of the candidate predictors (i.e., blinded)?
 The indices were apparent to the research team for death. Readmission data was collected blind to the candidate predictors. Readmission and death are regarded as objective outcome and the associated risk of bias is low. 2) Definition and methods for measurement of candidate predictors  The methods for measuring each index are provided, as well as the definitions of those in the PEARL score. A data collection guide was provided to each hospital which included definitions and guidance on data collection.
3) Timing of predictor measurement of candidate predictors (e.g. at patients presentation, at diagnosis, at treatment initiation)  Predictors were collected at the time of admission up to the point of the post-take ward round.

4) Were predictors assessed blinded for outcome, and for each other (if relevant)?
 The derivation cohort and external validation cohort were prospective. The internal validation cohort was performed retrospectively, but predictors were documented prior to the outcome (for example, eMRCD is collected within the COPD care bundle), and when extracting data reviewers were blind to the outcome. Collections of predictors were not blinded from each other, though the consequent risk of bias is low.

5)
Handling of predictors in the modelling (e.g. Continuous, linear, non-linear transformations or categorised)  Some predictors were categorised which is described.

Sample size 1) Number of participants and number of outcomes/events 2) Number of outcomes/events in relation to the number of candidate predictors (events per variable)
 1+2) In the derivation cohort, internal validation and external validation cohorts there were 824, 802 and 791 patients. There were 22 candidate predictors in the derivation cohort, and 309 events, or 14. events per index. The power calculation for the derivation cohort and each individual validation cohort is described.
Missing data 1) Number of participants with any missing value (include predictors and outcomes) 2) Number of participants with missing data for each predictor 3) Handling of missing data (e.g. complete-case analysis, imputation, or other methods)  1+2+3) Missing data by participant and by index provided. Missing data rates were low. Multiple imputation was used, and the approach and number of datasets used described. Five datasets were used which is regarded as sufficient given the amount of missing data; complete-case analysis was performed.  5) Shrinkage refers to adjusting coefficients to protect against over-fitting and loss of discrimination in validation studies. Weightings were assigned to the PEARL score based on the regression coefficients from the derivation study. There are various strategies that have been suggested to improve the generalisability of prognostic tools. Some of these do not apply when there are multiple validation cohorts. We took the mean regression coefficient across all cohorts, and compared it to those in the derivation study. "Previous admission" was under scored compared to the eMRCD score. We adjusted "previous admissions" from a weighting of 2 to 3. This fits with previous research that suggests previous admissions is consistently one of the strongest predictors of readmission risk, and will maximise the generalisability of PEARL. 2) Classification measures (e.g. sensitivity, specificity, predictive values, net reclassification improvement) and whether a priori cut points were used  Sensitivity and specificity are provided, with PEARL scores used as cut-offs. Reclassification measures, such as net reclassification improvement, look at the value in adding a single predictor to a prediction model. No reclassification measures were performed.

Model evaluation
1) Methods used for testing model performance: development dataset only (random split of data, resampling methods, e.g. bootstrap or cross-validation, none) separate external validation (e.g. temporal, geographical, different setting, different investigators)  Internal validation involved the same hospitals as the derivation cohort, but at a different time period (a form of temporal validation). External validation was performed at four hospitals. Hospitals were chosen for their differences, as described in the paper, to maximise generalisability. The research staff within external sites were not involved in the derivation or internal validation cohorts.
2) In case of poor validation, whether model was adjusted or updated (e.g. intercept recalibrated, predictor effects adjusted, or new predictors added)  Not applicable Results 1) Final and other multivariable models (e.g. basic, extended, simplified) presented, including predictor weights or regression coefficients, intercept, baseline survival, model performance measures (with standard or confidence Intervals) 2) Any alternative presentation of the prediction models. e.g. sum score, nomogram, score chart, predictions for specific risk subgroups with performance 3) Comparison of the distribution of predictors (including missing data) for development and validation datasets  1+2+3) Predictor weights and regression coefficients are given for the PEARL score. All models have AUROC calculated with confidence intervals. No subgroup analysis performed. Missing data rates for both all three cohorts was low.
Interpretation and discussion 1) Interpretation of presented models (confirmatory, if model useful for practice versus exploratory, is more research needed)  The performance of PEARL is shown in three cohorts, with consistent risk stratification. Quantifying the impact of using PEARL requires further research.
2) Comparison with other studies, discussion of generalisability, strengths and limitations.
 Described in discussion