To what degree can variations in readmission rates be explained on the level of the hospital? a multilevel study using a large Dutch database

• Published in: BMC health services research 2018-12, Vol.18 (1), p.999-999, Article 999
• Main Authors: Hekkert, Karin, Kool, Rudolf B, Rake, Ester, Cihangir, Sezgin, Borghans, Ine, Atsma, Femke, Westert, Gert
• Format: Article
• Language: English
• Subjects: Analysis; Comorbidity; Databases, Factual; Datasets; Health services; Healthcare quality indicator; Heart failure; Hospitals; Hospitals - statistics & numerical data; Humans; Logistic regression; Medical care; Multilevel analysis; Netherlands; Patient admissions; Patient readmission; Patient Readmission - statistics & numerical data; Pneumonia; Process Assessment, Health Care; Quality; Quality control; Quality Improvement - standards; Quality of Health Care - standards; Risk Factors; Software; Systematic review; Treatment Outcome; Variables
• ISSN: 1472-6963; 1472-6963
• DOI: 10.1186/s12913-018-3761-y

It is not clear which part of the variation in hospital readmissions can be attributed to the standard of care hospitals provide. This is in spite of their widespread use as an indicator of a lower quality of care. The aim of this study is to assess the variation in readmissions on the hospital level after adjusting for case-mix factors. We performed multilevel logistic regression analyses with a random intercept for the factor 'hospital' to estimate the variance on the hospital level after adjustment for case-mix variables. We used administrative data from 53 Dutch hospitals from 2010 to 2012 (58% of all Dutch hospitals; 2,577,053 admissions). We calculated models for the top ten diagnosis groups with the highest number of readmissions after an index admission for a surgical procedure. We calculated intraclass correlation coefficients (ICC) per diagnosis group in order to explore the variation in readmissions between hospitals. Furthermore, we determined C-statistics for the models with and without a random effect on the hospital level to determine the discriminative ability. The ICCs on the hospital level ranged from 0.48 to 2.70% per diagnosis group. The C-statistics of the models with a random effect on the hospital level ranged from 0.58 to 0.65 for the different diagnosis groups. The C-statistics of the models that included the hospital level were higher compared to the models without this level. For some diagnosis groups, a small part of the explained variation in readmissions was found on the hospital level, after adjusting for case-mix variables. However, the C-statistics of the prediction models are moderate, so the discriminative ability is limited. Readmission indicators might be useful for identifying areas for improving quality within hospitals on the level of diagnosis or specialty.