Mobile Edition
Print
Get the Mag
Weekly UpdatesAHRQ (2006) has published a complete description of the PSI measures, including how they are calculated. An important advantage in using the PSIs is that the software provided by AHRQ controls for variation in patient severity across hospitals by using the 3M Corporation's All-Patient Refined Diagnosis-Related Groups (APR-DRGs) method of risk adjustment. This method considers patient age, gender, diagnostic-related group, and potential comorbidities when calculating risk-adjusted rates. Most of the PSIs are risk adjusted similarly. The exceptions are death in low-mortality DRGs, foreign body left during procedure, and obstetric trauma from Cesarean delivery, which do not have a theoretical basis for risk adjustment (AHRQ 2006). Overall, the PSIs provide an important perspective on patient safety events that occur at a given institution, and they have been used extensively by researchers (Clement et al. 2007; Encinosa and Bernard 2005; Miller et al. 2001; Miller et al. 2005; Weiner et al. 2006; Zhan and Miller 2003).
Statistical Analysis
We used descriptive statistics to examine the distribution of each variable. Next, we calculated bivariate correlations with each of the variables of interest. To examine the relationship between IT adoption and the PSIs, we used linear regression analyses. Given the findings from our descriptive statistics, several PSI variables were log-transformed to facilitate analysis.
In a series of regression models, each IT measure was used as an independent variable to predict one of the PSIs as a dependent variable. Given their expected correlation (Burke and Menachemi 2004; Burke et al. 2002), the IT measures were specified in separate regression to avoid violating the assumption of multicollinearity. All models also controlled for several confounding factors. First, using data obtained from the Florida Agency for Health Care Administration, which we also used in our previous work (Menachemi et al. 2006; Menachemi et al. 2007a), we controlled for hospital financial performance. Researchers using data from Florida found that financial performance in hospitals was related to improved performance on select PSIs (Encinosa and Bernard 2005). Thus, a variable measuring total hospital expenses per day was included in each model. Previous studies have also suggested that hospital tax status (for-profit versus not-for-profit) is associated with both overall and clinical IT adoption by hospitals (Burke et al. 2002). Moreover, organization size can influence both IT adoption and other process and performance outcomes (Parente and Van Horn 2007; Warner, Menachemi, and Brooks 2005). Therefore, both tax status and bed size were included as control variables in each regression model.
RESULTS
Ninety-eight hospitals participated in the IT survey, representing a 49.5 percent response rate. All 98 hospitals were matched to their corresponding hospital discharge data. The organizational characteristics of responding hospitals are displayed in Table 1. Briefly, 70 hospitals (71.4 percent) were located in urban areas, 69 (72.6 percent) were not-for-profit, and 69 (71.1 percent) were part of a multihospital system. The average number of beds was 326, with a range of 15 beds to 1,862 beds. Hospitals adopted an average of 11.3 (45.2 percent) clinical IT applications, 15.7 (74.8 percent) administrative IT applications, and 5.0 (50 percent) strategic IT applications.
Overall, the average patient safety event rate of PSIs ranged from .00006 per 1,000 eligible discharges (for foreign body left in after procedure) to. 1541 per 1,000 eligible discharges (for obstetric trauma--vaginal with instrument).
In univariate analyses, most PSI outcome measures were not significantly correlated with any of the IT measures. However, with risk adjustment, complications of anesthesia was weakly correlated with clinical IT adoption (r = .245, p = .029), and failure to rescue was weakly and inversely correlated with clinical IT adoption (r = -.288, p = .008). In addition, risk-adjusted infections due to medical care was weakly correlated with clinical IT (r = .267, p = .013), administrative IT (r = .256, p = .017) and strategic IT (r = .341, p = .001) adoption; and risk-adjusted obstetric trauma (vaginal without instrument) was weakly correlated with clinical IT adoption (r = .328, p = .024).
In multivariate regression analyses, hospitals that adopted a greater number of IT applications were significantly more likely to have desirable quality outcomes on eight PSI measures. For example, an increase in the total number of clinical IT applications was significantly inversely correlated with death in low-mortality DRGs ([beta] = -1.82, p = 0.024), risk-adjusted rates of decubitus ulcer ([beta] = -1.30, p < 0.001), and risk-adjusted postoperative sepsis ([beta] = -1.69, p = 0.031).
An increase in clinical IT was further associated with two additional PSIs--risk-adjusted postoperative hemorrhage and risk-adjusted postoperative pulmonary embolism--but the trends ([beta] = -1.33, p = 0.060 for postoperative hemorrhage and [beta] = -.977, p = 0.064 for postoperative pulmonary embolism) did not reach statistical significance. Nevertheless, an increase in the number of administrative IT applications was associated with lower rates of risk-adjusted decubitus ulcer ([beta] = -2.49, p < 0.001).
An increase in the adoption of strategic IT applications was significantly inversely related to six PSI measures. Hospitals reporting the adoption of additional strategic IT applications had more desirable risk-adjusted rates of selected infections due to medical care ([beta] = -1.00, p = 0.044), postoperative hip fracture ([beta] = -2.42, p = 0.013), postoperative respiratory failure ([beta] = -2.32, p = 0.025), postoperative sepsis ([beta] = - 1.75, p = 0.015), postoperative wound dehiscence ([beta] = -2.09, p = .037), and accidental puncture or laceration ([beta] = -.135, p = 0.05).
Note that no relationship identified in multivariate analyses was in the unexpected direction. In other words, any time that a significant relationship was identified between a PSI outcome and a measure of IT adoption, the relationship was in the anticipated (desirable) direction.
DISCUSSION
A growing body of literature has linked the use of IT to improved patient safety (Bates et al. 1998; Johnson et al. 2002; Kaushal and Bates 2002; Overhage et al. 1997; Poon et al. 2006). Nevertheless, a large number of such studies cannot be generalized to typical hospitals because they were conducted in specialized academic settings with unique circumstances (Chaudhry et al. 2006). Only a small number of studies have considered the entire portfolio of IT capabilities in a set of hospitals, and they found links between IT and improved financial performance (Devaraj and Kohli 2000; Menachemi et al. 2006; Wang et al. 2005), operational performance (Bhattacherjee et al. 2006; Parente and Van Horn 2007), and nonclinical outcomes such as perceptions of quality (Rodger, Pendharkar, and Paper 1999) and attitudes about employee satisfaction (Hatcher 1998).
We explored the relationship between IT adoption and performance on a set of widely used patient safety indicators in a relatively large sample of hospitals. We hypothesized that hospitals with greater IT adoption would perform better on patient safety measures, and our findings support this notion. Eight different PSI measures were statistically significantly related to at least one measure of IT adoption. Moreover, all significant relationships were in the negative (desirable) direction, suggesting that increases in IT adoption are associated with a decrease in adverse events.
We also hypothesized that hospitals with more sophisticated IT infrastructures would perform best on the set of PSI measures. Indeed, greater adoption of strategic information systems--the proxy measure for overall IT sophistication--was significantly related to the greatest number of patient safety outcomes. Many of the outcomes, including selected infections due to medical care, postoperative respiratory failure, and postoperative sepsis, are conditions that can be prevented when clinicians have access to up-to-date patient information, standardized order sets, and evidenced-based guidelines. Hospitals with sophisticated and integrated information systems are able to ensure that clinicians receive critical clinical information at the point of care and to assist physicians in adhering to proven clinical guidelines (Bates et al. 2001; Bates and Gawande 2003; Schiff et al. 2000; Schiff et al. 2003).
Our analyses and research protocol have several strengths. First, we used validated and widely used measures of IT and patient safety. Second, by combining primary and secondary data, we were able to overcome common method bias, a methodological problem inherent in many single-respondent surveys. Third, we conducted instrument construction, sampling, and statistical analyses using a systematic, rigorous, and scientific approach. Thus, our data were expected to generate findings that were strong in internal validity. Moreover, our approach toward IT measurement did not focus on a specific IT vendor or product. Instead, we examined the homogenous effect of IT by studying a heterogeneous sample of hospitals.
When interpreting our results, some research limitations should be considered. First, our study had an observational cross-sectional design that is not suitable for detecting causality. We merely identified a significant relationship between IT adoption and improved patient safety outcomes. An unobserved variable (e.g., effective management) might be associated with both IT adoption and superior patient safety in hospitals and, thus, affect the results presented. Other limitations of the study include the fact that our data were based on 98 hospitals, representing a response rate of approximately 50 percent. Although the sample of hospitals is relatively large, especially compared to previous studies of this kind, and the participation rate is acceptable, we recognize that response bias is a possibility. Finally, by design our study focused exclusively on acute care hospitals in one state. As such, any attempt to generalize our findings to other types of healthcare settings or to acute care hospitals in other states must be undertaken with caution.
FEED