Seasonal mortality trends for hospitalised patients with acute kidney injury across England

Study designs

This was a retrospective observational study of adult hospitalised patients (> 18 years) with AKI in England, between 01/01/2017–31/12/2017, at acute healthcare organisations (termed ‘trusts’ in the NHS). Each NHS Trust may consist of between one and five acute hospitals, with some shared infrastructure e.g. laboratory information management systems (LIMs).

Dataset

Laboratories in England are mandated to monitor for significant elevations in serum creatinine using the national AKI alerting algorithm published by NHS England. [14] AKI alerts for AKI stages 1, 2 and 3 (graduated by level of severity) are sent by laboratories to the treating clinicians and also the UKRR. This database of laboratory returned biochemical AKI alerts held at the UKRR is known as the AKI Master Patient Index (MPI).

AKI episodes and clinical settings are defined as per the 2020 UKRR AKI report. [6] The date of a first AKI episode is defined as the date of the first AKI alert received by the UKRR for each individual patient. Subsequent alerts are only considered to be a further episode of AKI if at least 30 days have passed since the last alert. If an episode appears to last more than 90 days, duration of the episode is truncated to day 90 to align with the KDIGO definition of chronicity after 90 days of an AKI episode. [6,15]

Study cohort

The UKRR routinely link data for patients in the AKI-MPI to NHS Hospital Episode Statistics (HES) and mortality feeds from the Office for National Statistics (ONS). In this analysis, we studied patients in the dataset between 01/01/2017–31/12/2017. Appendix A1a and A1b outlines further details on cohort selection.

Exposure

Seasons were defined meteorologically as spring (March, April, May), summer (June, July, August), autumn (September, October, November) and winter (December, January, February).

Outcome

The main outcome of interest was mortality in hospital or within 30 days of discharge.

Covariates

Clinical variables included in our model were primary diagnosis at hospital admission (see appendix A2), comorbidity RCCI (AKI specific re-weighting of Charlson Comorbidity Index as detailed in Appendix table A1c ), age, sex, admission method, peak AKI stage, index of multiple deprivation quintiles (IMD) derived from patient postcode, ethnicity, whether AKI was community acquired prior to hospital admission (CAH) or hospital acquired (HA). [16].

Statistical modelling

All analyses were conducted using SAS 9.4. Proportions of patients with different clinical characteristics across the seasons were presented. Chi-square tests were performed to see if patient characteristics have an association with season. P values less than 0.05 were considered statistically significant.

Multivariable logistic regression was used to assess whether season is associated with the 30-day mortality, adjusting for clinical variables. This paper aims to demonstrate the net effect of season to mortality of AKI patients. Since we are not focused on the mortality prediction, or interpreting each of the covariate effects on mortality, we have not shown the individual point estimates for all the covariates. We did not model any interactions as we were concerned about overfitting models and spurious associations. Age was found to act as a non-linear covariate and was therefore included as a non-linear cubic spline (Appendix A1d). Risk factors were entered into the model in each forward step sequentially, allowing assessment of the impact of each risk factor on 30-day mortality and whether they were confounded by other risk factors already in the model. The model was internally validated using Hosmer and Lemeshow measure of concordance. The concordance statistic of the model was 77%, indicating acceptable goodness of fit.

Season was also modelled by centre to explore how seasonal mortality trends varied across NHS Trusts. The ‘Akaike Information Criterion (AIC)’ was used for variable selection and for assessing the inclusion of interactions between season and centre. [17].

留言 (0)

沒有登入
gif