Categories: Featured Articles, News, Newsletter Issue 2018:1


Information Technology and Patient Health: Analyzing Outcomes, Populations, and Mechanisms

by Seth Freedman, Haizhen Lin, & Jeffrey Prince

Electronic Medical Records (EMRs) have become a widespread input throughout the healthcare delivery system. At the outset of this diffusion there was substantial optimism that these tools would lead to improvements in productivity, costs, and quality. This optimism was instrumental in the passage of the Health Information Technology for Economic and Clinical Health (HITECH) Act in 2009, which included large financial incentives for EMR adoption and usage. In contrast, the empirical literature measuring the impacts of EMR adoption on patient health outcomes has not found strong evidence of improvements. However, these studies have largely focused on mortality among Medicare patients, with the exception being one study that found EMR diffusion decreased infant mortality rates (Miller & Tucker 2011).

In this paper we estimate the effect of hospital EMR adoption on health outcomes using data that allow us to capture impacts for a population that includes not only Medicare patients, but other adult patients as well. In addition, we focus on preventable adverse health events, which are important patient outcomes but less extreme than mortality. We also explore heterogeneity by both patient characteristics and EMR types with the goal of detecting which mechanisms incorporated into EMR systems may be most effective.

We use annual data on hospital EMR adoption from the Healthcare Information and Management Systems Society (HIMSS) Analytics Database for the years 2003 through 2010. The HIMSS survey collects a wide range of information on a hospital’s health information technology applications, and we focus on two indicators of “advanced” EMR adoption, namely Computerized Physician Order Entry (CPOE) and Physician Documentation (PD). CPOE could reduce adverse events in hospitals by reducing errors and miscommunications in the ordering process. In addition, while decision support applications are not well measured in our data, most CPOE systems interact directly with decision support tools that can provide error checking, diagnostic or therapeutic protocols, or other clinical guidelines. PD could reduce adverse events through better data accessibility, increased legibility, and accuracy. In addition, PD can support more effective communication and coordination when care is administered by multiple practitioners.

We merge our measures of EMR adoption to nationally representative hospital discharge data from the Nationwide Inpatient Sample (NIS). Each year, the NIS creates a 20-percent, nationally representative, stratified sample of U.S. community hospitals, and obtains visit level data from all discharges that year. For each record, we observe diagnosis, treatment, and demographic information, and using tools from AHRQ we construct indicators for discharges that suffered various Patient Safety Indicators (PSIs), which are preventable, in-hospital, adverse events.  We identify a subset of these PSIs that are likely to be impacted by EMR adoption, because they can be prevented by improving decision making and oversite processes. We also identify a subset that are unlikely to be impacted by EMR adoption as a set of falsification outcomes, because they are mostly a function of physician skill or physical accident. While we do not observe each hospital every year in the NIS, we observe many of them at multiple points in time, allowing us to use within hospital variation in EMR adoption over time in our empirical analysis.

We estimate discharge-level linear probability models that include hospital and time fixed effects. The dependent variables in our regressions are indicators for whether a patient experiences a PSI. In our main estimates, we aggregate the PSIs for which we expect EMRs to have some effect into a measure of whether or not the patient experiences at least one postoperative adverse event. The key independent variables are indicators for EMR adoption and an interaction between EMR adoption and an indicator for whether the patient is a “simple” case, to explore patient heterogeneity. We consider various different definitions of simple cases, including those with few comorbidities, common diagnoses, low mortality risk, and low functional severity.

In our main estimates we find that along all dimensions of case complexity, CPOE adoption leads to statistically significant declines in postoperative adverse events for simple cases, but has no statistically significant effect on complex cases. For example, CPOE decreases postoperative adverse events by 11% of the sample mean for patients with few comorbidities, but has no impact for those with many comorbidities. CPOE decreases adverse events by 17% for patients with common diagnoses, but has no impact for those with less common diagnoses. We find similar signed effects for physician documentation, but the point estimates are smaller and are not statistically significant for any of the subgroups.

In addition to our main fixed effect estimates, we estimate event study specifications. These specifications allow us to both confirm that there were no differential trends in adverse events in the pre-adoption periods and to assess how effects evolve over time. We find consistent evidence that among both complex and simple cases, there are no pre-adoption changes in adverse events. For physician documentation, we find no evidence of effects in any post-adoption years. For CPOE, we find that the decrease in adverse events for simple patients tends to occur one year after the adoption year and persist into future years.

To further assess our identification strategy, we show that observed patient characteristics did not change after EMR adoption, including the composition of patients between simple and complex case types. We also find that there is no impact of EMR adoption on our falsification outcomes. Neither CPOE or physician documentation impact PSIs related to surgical skill or physical accident, suggesting that our results are not driven by other interventions occurring in hospitals to improve patient safety concurrently with EMR adoption.

Taken together, our findings suggest that CPOE, but not physician documentation, improves patient safety among less complex cases. In our paper we postulate that this pattern of results is suggestive of decision support mechanisms. Decision support tools often tied to CPOE are likely to be most impactful for preventing adverse events among relatively “simple” patients, because current decision support systems are most useful in cases with clear guidelines and protocols rather than cases with multiple interacting conditions and other complexities. On the other hand, care coordination aspects of PD are likely to be more impactful for “complex” patients, consistent with findings by McCullough et al. (2013) that EMRs improve mortality for the highest severity Medicare patients, particularly those needing multiple coordinated providers or synthesis of large amounts of clinical information. Our results therefore complement this past literature by pointing out that different mechanisms can improve care along different metrics of quality and for different patient populations.

References

McCullough, Jeffrey S., Stephen Parente, and Robert Town. 2013. “Health Information Technology and Patient Outcomes: The Role of Organizational and Informational Complementarities.” NBER working paper no. 18684.

Miller, Amalia R. and Catherine E. Tucker. 2011. “Can Health Information Technology Save Babies?” Journal of Political Economy 119(2): 289-324.