Implementing EHR in healthcare settings incurs challenges in maintaining quality care, particularly in terms of efficiency, timeliness and safety of care. There are limited studies describing best practices in avoiding reduction in the quality of care.^{4 5} As EHR systems evolve from first generation computerized physician order entry or simple niche systems to large hospital-wide EHR integrated records, the risks of the implementation on impacting patient safety, efficiency and financial recovery increase dramatically.^6–8 Given the breadth of potential negative impact, the ‘success’ of any implementation is colored by one's perspective, whether practitioner, administrator, or patient, resulting in little consistency in the definition of a ‘successful’ implementation.⁹ A variety of models in different healthcare settings managing workflow and expectations has been theorized, but there is little evidence to support any one of these models during implementation.¹⁰ It is clear that strategies employed in one setting (such as decreased scheduled visits, postponing elective surgeries, increasing office hours) cannot always be translated to another.

The emergency department (ED) faces unique challenges in an EHR implementation. EDs at baseline are an environment in which multiple barriers naturally exist for negatively impacting quality care, including job stress, limited staffing, multiple interruptions and most important, crowding. ED crowding is a well-described problem within our healthcare system, with many contributing factors, which has been shown to affect quality negatively.¹¹ Even though limited data have shown that efficiency can improve long term after an EHR implementation, provider buy-in with the ability to operate safely during the transition is a very large factor in the success of an implementation in this setting.¹² Providers in the ED environment work to maintain quality, and any potential negative impact of quality will be negatively viewed, EHR implementation being a good example. National benchmarking for quality in the ED setting places significant focus on timeliness, safety and efficiency standards, most directly reported as overall length of stay (LOS), door to doctor times and left without being seen, using them as proxies for quality measures that are difficult to collect, such as time to analgesia, antibiotics, and patient satisfaction.¹³ To date, there have been no studies that describe best practices to mitigate reduction in quality care measures in the ED during EHR implementation.

Our study describes our experience with one potential intervention to decrease delays in ED care during EHR implementation. The strategy employed was modeled after our institution's response to H1N1, an overflow clinic separate from the ED to care for low acuity patients.^14–16 Before H1N1 we were scheduled to implement our EHR within the time frame of the expected surge of patients. Rather than delay rollout of our EHR in the ED, we opted to use the strategies developed during the H1N1 response during our EHR implementation.

Table 1 outlines the 2-week time blocks and reports the patient flow data throughout the timeline. Each 2-week block was similar in terms of the number of patients seen (mean 3281, range 3154–3333), and numbers of discharged and admitted patients. When compared with pre-implementation levels, on average, admitted patients' LOS was 6–20% longer during EHR implementation and for discharged patients, LOS was 12–22% longer.

During the overflow clinic activation, and before EHR implementation nearly 10% of patients were diverted to the overflow clinic; however, only 5% were diverted during EHR implementation.

The overflow clinic during H1N1 surge did appear to reduce the overall LOS before the implementation (95% CI showed decreased LOS 24 to 53 min for admissions, and by 9 to 19 min for discharges with the clinic in place, comparing time 1 with time 2). However, during EHR implementation, the overall LOS for both groups exceeded both the H1N1 pre-overflow clinic block as well as the H1N1 overflow clinic block (95% CI showed increased LOS 32 to 62 min for admissions and by 35 to 44 min for discharges compared with the clinic in place before rollout). Both LOS dropped back to pre-H1N1 average LOS after 3 months.

Figure 2 reveals that it took approximately 3 months to get to the previous steady state level for visit metrics, and a new improved LOS steady state was achieved approximately 6 months post-implementation. Using the new steady state data to calculate proportions, the ED system operated at 80% efficiency in terms of LOS during the first 2 weeks of implementation, and recovered to 90% by weeks 3–4. This included additional patients who were diverted to an overflow clinic. Despite consistency during all blocks in time to room placement, there was a significant increase in time to doctor in the first 2 weeks after EHR implementation from 47 min to 70 min (p<0.001).

Table 2 shows the total number of clinical hours worked by all medical care providers in our ED by month. Despite a consistent total number of clinical work hours from October to December 2009, the patients/h seen fell between October and December as the total number of patient visits dropped. Staffing levels remained elevated the following year compared with 2009, resulting in an overall lower patients/h seen in the last quarter of 2010 (total provider hours increased by 6% and total patients decreased by 16%).

Using the model of input/throughput/output to describe ED flow in our setting,¹¹ the input (number of visits, time to room placement) and output (numbers of admitted and discharged patients) measures were consistent between each of the 2-week blocks. Given that no other interventions were employed either in the ED or within the hospital during this time (other than increasing staffing in the ED), we conclude that the impact for delays in care were directly related to the implementation of EHR.

The difference in LOS was seen in the time to provider, and was not reflected in time to room placement. Time to room placement is a proxy for the amount of time doing rapid nursing assessments and triage. Before implementation, there was a concern that this process, designed to take less than 10 min per patient, would be affected greatly by the new EHR, but that was not the case. All staff experienced the same change in system, but more ‘bottlenecks’ appeared to be attributable to provider slowdown. This is clearly in keeping with previous research, in which loss of physician efficiency was recognized to be a potential drawback to EHR systems.⁹ This effect might be more pronounced in our setting, where the frequent rotation of residents (involved in approximately 50% of visits) increased ‘new’ provider users to the system. Whether this is offset by improving revenues with increased charting or capturing of charges was not addressed by our study.

A critical question was whether the overflow clinic model could be quickly adapted to offload the ED for the implementation of EHR. The clinic was successful at decreasing LOS during our flu surge, but it was not effective in diverting patients during EHR implementation, perhaps partly due to fewer patients being diverted to the clinic. The clinic was designed to divert patient with ‘flu’ symptoms, and once that specific population declined, the process for triaging patients to the clinic did not work well, and fewer patients were diverted. It is unclear from our data whether increasing the specific number of patients diverted would have made a difference in decreasing LOS during the implementation.

We saw continued improvement in LOS from 3 to 6 months after implementation, leading to an overall improvement compared with baseline. It appears from productivity data that the total number of hours worked by all providers has remained constant since H1N1 and implementation, despite an overall reduction in patient visits, which probably explains this continued drop. The drop in total ED visits was unexpected in the following year, and probably impacts the perceived loss of provider ‘productivity’ seen 1 year after implementation.

Provider productivity was included to determine how much an effect was due to increasing provider and patient services staffing during the go live. Initial decreased productivity seen in December 2009 could be due to the product, but it is interesting that this effect was not seen in the month of implementation. Due to the structuring and reporting of these data, it was not possible to segment it further.