Does Hospital Infection Reporting Affect Actual Infection Rates, Reported Rates, or Both?: A Case Study of Pennsylvania

Background: Health-care associated infections (HAIs) kill about 100,000 people annually; many are preventable.  In response to this epidemic, 15 states require hospitals to publicly report their infection rates and national reporting of central-line associated bloodstream infection (CLABSI) is planned.  Yet the effect of public reporting on hospital behavior, and thus on infection rates is unknown.

Objective: We examine two broad questions.  The first involves reputation markets and whether public reporting of quality measures leads to higher quality.  The second involves the incentives of those subject to public reporting to game the reporting rules.

Methods: Our key innovation is to study both the directly reported CLABSI rate, available on the Pennsylvania Health Care Cost Containment Council (PHC4) website for 2005-2007, and a related proxy infection rate, derived from hospital inpatient "billing" records for 1998-2008  ("inpatient CLABSI rates"), which is not publicly reported, and hence not directly subject to gaming incentives.  Two sources of inpatient data are 1) Pennsylvania Inpatient Data and 2) Nationwide Inpatient Sample for ten non-reporting states ("control states").  Using inpatient CLABSI rates in both Pennsylvania and control states, we conduct difference-in-differences (DiD) analysis to identify the impact of public reporting.  By comparing inpatient rates to publicly reported rates, we assess whether hospitals appear to be gaming the public rates.

Results: We find evidence for both effects.  First, public reporting appears to affect actual infection rates.  Inpatient CLABSI rates show similar trends in Pennsylvania and control states during the pre-reporting period of 1998-2004.  Pennsylvania diverges from the control states during the reporting period of 2004-2008; with inpatient rates dropping on average by 21% in Pennsylvania, versus negligible change in the control states.  This estimate is reliable because hospitals do not publicly report inpatient rates and thus have no incentive to manipulate them.  Hospitals with high publicly reported rates in 2005 (> 0.40/1,000 discharges) experience drops by 2007; the higher the initial rate, the larger the decline.  Hospitals with lower 2005 rates do not.   Second, public reporting appears to induce at least some gaming.  Publicly reported infection rates drop by 40% over 2005-2007, far faster than inpatient rates; the correlation between the two rates drops over this period, and the correlation between inpatient rate changes and public reporting changes is surprising low.

Conclusions: Public reporting appears to cause the relative decline in Pennsylvania CLABSI rates. However, evidence of gaming confirms the value of measuring the impact of public reporting on quality using non-public measures (such as our inpatient rate), and offers reasons to be skeptical about studies that rely solely on reported measures.