Background: CMS and other entities have focused on reduction of readmissions as a national quality improvement goal. The 30-day risk standardized readmission rate is derived from administrative data and requires the accurate coding of a principal diagnosis. We sought to validate the coded principal diagnosis for patients admitted with pneumonia and to determine if local efforts to improve inpatient coding improved data validity.
Methods: Two attending physicians reviewed all index admissions with a principal diagnosis of pneumonia from Fiscal Years (FY) 2014-2015 at a large, tertiary care teaching hospital using a standardized abstraction form. Index admissions that did not meet the data definition for the 30-day risk-standardized readmission rate were excluded. Each reviewer independently determined a principal diagnosis for each admission. Disagreements between reviewers were resolved by consensus when possible, or by a third reviewer when needed. In FY 2015, the Michael E. DeBakey VA Medical Center implemented a comprehensive intervention to improve medical documentation and coding. The bundle included resident and attending education and twice-weekly coding huddles with physicians and coders. The proportion of admissions with a confirmed principal diagnosis of pneumonia or alternative diagnosis were compared before and after the coding intervention using a chi square test, and kappa statistics were used to measure interrater reliability.
Results: In FY 2014, 156 admissions met the inclusion criteria. Of these, 29 (18.6%) admissions had a principal diagnosis of pneumonia on chart review. The remaining 127 (81.4%) admissions were determined to have a more appropriate alternative principal diagnosis, most commonly sepsis (n=88) and acute hypoxic respiratory failure (n=20). In FY 2015, only 92 admissions met the inclusion criteria likely due to an increase in the use of observation status. Of these, 18 (19.6%) admissions had a principal diagnosis of pneumonia on chart review. The remaining 74 (80.4%) admissions were determined to have an alternative principal diagnosis (sepsis n=51, acute hypoxic respiratory failure n=17). The intervention did not change the proportion of admissions with a confirmed coded principal diagnosis of pneumonia (p=0.85). Interrater reliability was moderate (kappa = 0.42).
Conclusions: We found that less than one fifth of admissions were correctly categorized with a principal diagnosis of pneumonia. This proportion did not improve meaningfully with a coding intervention. Our findings call into question the validity of data used in reporting 30-day readmission rates. The disagreement between experienced physicians regarding the appropriate principal diagnosis highlights the challenges clinicians face in clinical documentation. Recent data definition changes from CMS to include a principal diagnosis of sepsis with a secondary diagnosis of pneumonia as a pneumonia admission may improve the validity of the data.