The American Journal of Managed Care April 2011
Electronic Health Record Functions Differ Between Best and Worst Hospitals
Objective: To determine whether patterns of electronic health record (EHR) adoption and “meaningful use” vary between high-, intermediate-, and low-quality US hospitals.
Study Design: We used data from the Hospital Quality Alliance program to designate hospitals as high quality (performance in the top decile nationally), low quality (bottom decile), and intermediate quality (all others). We examined EHR adoption and meaningful use using national survey data.
Methods: We used logistic regression models to determine the frequency with which hospitals in each group adopted individual EHR functions and met meaningful use criteria, and factor analyses to examine adoption patterns in high- and lowquality hospitals.
Results: High-quality hospitals were more likely to have all clinical decision support functions. High-quality hospitals were also more likely to have computerized physician order entry for medications compared with intermediate- and low-quality hospitals. Among those who had not yet implemented components of clinical decision support, two-thirds of low-quality hospitals reported no concrete plans for adoption. Finally, high-quality hospitals were more likely to meet many of the meaningful use criteria such as reporting quality measures, implementing at least 1 clinical decision support rule, and exchanging key clinical data.
Conclusions: We found higher rates of adoption of key EHR functions among high-quality hospitals, suggesting that high quality and EHR adoption may be linked. Most low-quality hospitals without EHR functions reported no plans to implement them, pointing to challenges faced by policy makers in achieving widespread EHR adoption while simultaneously improving quality of care.
(Am J Manag Care. 2011;17(4):e121-e147)
For hospitals seeking to improve care, focusing on specific electronic health record (EHR) functions, particularly order entry with clinical decision support, is likely a key part of achieving high-quality performance.
- Implementing clinical decision support functions in low-quality hospitals should be a goal for policy makers.
- High-quality hospitals tend to cluster results viewing and clinical documentation functions, a unique adoption pattern that may serve as an effective adoption model.
- Widespread resistance to EHR adoption, especially among low-quality hospitals, points to challenges ahead as policy makers seek to ensure that all Americans receive high-quality hospital care.
Since the passage of the Health Information Technology for Economic and Clinical Health (HITECH) Act, several studies have called into question the relationship between EHR use and quality of care.8,9 These data have fueled criticisms of current efforts to promote EHR adoption; skeptics point to these studies to argue that there is inadequate evidence to support widespread EHR use. However, studies demonstrating only modest overall effects of EHRs on quality of care may miss important differences in EHR use between the best and worst hospitals. If the underlying goal is to improve quality, examining how high-quality hospitals in the United States use EHRs and determining whether this is substantively different from how poor-quality hospitals use EHRs could provide important insights for clinicians and policy makers seeking to move providers toward the provision of higher quality care. Further, understanding which specific EHR functionalities are in use among the high-quality hospitals could provide guidance in terms of how low- or intermediate-quality hospitals might focus their EHR efforts going forward.
We used national data on patterns of EHR adoption to address 4 key questions. First, were there differences in the adoption of specific EHR functionalities (eg, medication lists, computerized prescribing, clinical decision support) between high- and low-quality hospitals? Second, if these dif-ferences exist, which functionalities displayed the largest disparities in adoption when comparing high- and low-quality hospitals? Third, did the highest quality hospitals seem to have different patterns of adoption than the lowest quality hospitals (ie, did the cluster of functions adopted vary between the high- and low-quality institutions)? Fourth, among those hospitals that have not yet adopted individual functionalities, were there important differences between highand low-quality hospitals in their current plans to implement them? And finally, were there differences in adoption of the specific functions that comprise the newly established meaningful use criteria10 for EHR adoption?
Measures of Electronic Health Record Functions
We used 2 primary data sources for this analysis: the 2009 American Hospital Association (AHA) hospital information technology (IT) survey of US acute care hospitals and the 2006 Hospital Quality Alliance database. The AHA IT survey was distributed as a supplement to the AHA’s annual survey in 2009. This survey has served as a data source for many analyses, and the details of its development and distribution are described in prior publications.10 The survey was administered to all 4493 acute care hospitals in the AHA (an estimated 97% of all hospitals in the United States) from March through September 2009. A total of 3101 surveys were completed, for a 69% response rate. The survey assessed the level of adoption of specific EHR functionalities. Respondents were asked to report a score of 1 through 6 to assess the degree of adoption for each functionality, ranging from full adoption of the function across all units to a declaration that the functionality was not in place and that there were no plans or considerations to implement it. We focused on the 24 electronic functions that a federally sanctioned expert panel identified as part of a comprehensive EHR.10
Measures of Quality
We used data from the Hospital Quality Alliance, which contains information on process measures for patients cared for during calendar year 2006. We created summaryscores for performance on care for acute myocardial infarction, congestive heart failure, pneumonia, and prevention of surgical complications.11 The specific indicators are summarized in Appendix 1. We took an average of each hospital’s summary score within each of the 4 clinical areas and ranked all the hospitals in order of performance. We excluded hospitals with fewer than 30 observations for any of the 4 clinical conditions of interest, as well as hospitals located outside of the 50 states and the District of Columbia.
We began by categorizing the hospitals in our sample into quality deciles based on their overall quality score and created 3 groups for our main analysis: hospitals in the top 10% of performance were designated as high quality, those in the bottom 10% were designated as low quality, and all other hospitals (those in deciles 2 through 9) were designated as intermediate quality. In sensitivity analyses, we examined other cut-points for designating hospitals as high versus low quality, including the top and bottom 20% as well as top and bottom 30%. We calculated the proportion of hospitals within each cohort (high quality, medium quality, and low quality) that had adopted each EHR functionality in at least 1 hospital unit. We used X2 tests to compare the proportions of hospitals that had adopted each function across the 3 groups. To account for potential confounding, we built multivariate logistic regression models, adjusting for hospital size, region, ownership (for profit, nonprofit, or public), teaching status, membership in a hospital system, urban versus nonurban location, presence of a cardiac intensive care unit (an indicator of technological capacity), and the percentage of each hospital’s patients who were covered by Medicaid (an indicator of the socioeconomic status of patients treated in each hospital). For each specific functionality, hospitals with missing data were excluded from that calculation. We only included the presence of several key decision support tools related to medication alerts if the hospital also had computerized provider order entry (CPOE) for medications. This was done to reflect true decision support at the point of care by healthcare providers, which would require the presence of electronic order entry. We reran our analyses without the requirement for CPOE and our results were qualitatively similar. Thus, we only present the findings of those decision support tools in the presence of CPOE.
Next, we used factor analysis to determine the covariance of adoption of functionalities within each of the quality cohorts. We simply describe the patterns of clustering of functions across the 3 quality cohorts.
Using the same groups but limiting our analysis this time to those hospitals that had yet to implement each EHR functionality, we calculated the proportion of hospitals that reported no concrete plans for implementation (defined as the proportion reporting either that they had considered but had no resources identified for EHR implementation or that they had no plans to implement EHRs). We compared the frequency of these responses across the 3 groups initially using X2 tests and subsequently using multivariate logistic regression analyses as described above to adjust for potential confounders.
Finally, we examined the proportion of hospitals within each quality cohort that had adopted the specific functions required to meet meaningful use criteria. These included 12 objectives that had clear analogues to the AHA health IT survey (9 of the 14 Core Objectives and 3 of the 10 Menu Objectives; see Appendix 2). For these analyses, we used X2 tests to determine whether the proportion of adopters varied across these 3 groups and did not exclude missing data from calculations.
There were slight differences between hospitals that did and did not respond to the health IT survey.10 In the analyses reported, all results were weighted to account for the differences due to nonresponse using a previously described method. 10 All analyses were performed using Stata/SE, version 10.1 (StataCorp, College Station, TX). A 2-sided P value less than .05 was considered to be statistically significant.
Of the 1637 hospitals in our sample, 166 were designated as high quality, 1318 as intermediate quality, and 153 as low quality (Table 1). There were substantial differences in the characteristics of these hospitals: high-quality hospitals were more often large compared with low-quality hospitals (26% vs 8%, P <.001), and more often nonprofit in ownership (84% vs 49%, P <.001). High-quality hospitals were significantly more likely than low-quality hospitals to be teaching hospitals (44% vs 23%, P <.001), to belong to a hospital system (71% vs 55%, P <.05), to be located in urban areas (86% vs 59%, P <.001), and to have a dedicated coronary intensive care unit (62% vs 28%, P <.001). Finally, the percentage of patients with Medicaid was substantially lower in the highquality than the low-quality hospital cohort (9% vs 15%, P <.001).
We found substantial differences in the adoption of EHR functions among the 3 groups of hospitals (Table 2). Highquality hospitals more often had electronic nursing notes (81% vs 73% and 68%, P = .04) and medication lists (89% vs 79% and 73%, P <.01) than intermediate-quality and lowquality hospitals, respectively. All decision support tools had significantly higher adoption levels in the high-quality cohort. The differences between the high- and low-quality cohorts in adoption of all of these clinical decision support functions ranged from 17% to 20%, and all were significant (Table 2).
After multivariable adjustment, we found that adoption of 22 of the 24 functions was still higher in high-quality hospitals, although most of the differences were no longer statistically significant (Appendix 3). Functions for which the differences across the 3 quality cohorts were statistically significant included problem lists, medication lists, diagnostic test images, and many of the clinical decision support tools.
In sensitivity analyses, when we examined groupings based on alternative cut-points, we found that most of the results were qualitatively similar. However, expanding the high- and low-quality groups to the 30% cutoff decreased the differences between groups, some of which became nonsignificant (see Appendix 4).