Clinical impact of intraoperative electronic health record downtime on surgical patients

Abstract

Objective

Despite increased use of electronic health records (EHRs), the clinical impact of system downtime is unknown.

Materials and Methods

This retrospective matched cohort study evaluated the impact of EHR downtime episodes lasting more than 60 minutes over a 6-year study period. Patients age 18 years or older who underwent surgical procedures at least 60 minutes in duration with an inpatient stay exceeding 24 hours within the study period were eligible for inclusion. Out of 4115 patients exposed to 1 of 176 EHR downtime episodes, 4103 patients were matched to an unexposed cohort in a 1:1 ratio. Multivariable regression analysis, as well as trend analysis for effect of duration of downtime on outcomes, was performed.

Results

Downtime-exposed patients had operating room duration 1.1 times longer (p < .001) and postoperative length of stay 1.04 times longer (p = .007) compared to unexposed patients. The 30-day mortality rates were similar between these groups (odds ratio 1.26, p > .05). In trend analysis, there was no association between duration of downtime with respect to evaluated outcomes, postoperative length of stay, and 30-day mortality.

Conclusion

EHR downtime had no impact on 30-day mortality. Potential associations for increased postoperative length of stay and duration of time spent in the operating room were observed among downtime-exposed patients. No trend effect was observed with respect to duration of downtime and postoperative length of stay and 30-day mortality rates.

INTRODUCTION

Following the enactment of the Health Information Technology for Economic and Clinical Health (HITECH) Act,¹ there was a 9-fold increase in electronic health records (EHRs) adoption from 2008 through 2015.² In the same time period, the proportion of hospitals using comprehensive EHRs continued to rise, in contrast to the use of basic EHRs, which offer limited functionality.² This increase in use and functionality has been accompanied by a growing body of literature^3–5 that addresses the unintended adverse consequences of EHR use.

Overdependence on technology has been identified as 1 of 9 major types of unintended adverse consequences in the evaluation of computerized order entry (CPOE).⁶ In 1 follow-up study system downtime has been included in this category.⁷ Downtime describes instances where an electronic system does not function optimally, due to either an impairment or outage. According to surveys of large health care systems in the United States, almost all respondents reported experiencing downtime, with a majority experiencing at least 1 downtown of more than 8 hours.⁸ In May 2017, the effects of the WannaCry ransomware cyberattack on the UK National Health Service (NHS) were reported by popular press as preventing providers from accessing their computers.^9,10 Several NHS trusts were affected which resulted in the cancellation of appointments and surgeries and the diversion of emergency health care.¹¹ While data related to the national incidence of EHR downtime are lacking, the problem appears common.

The objective of this study was to evaluate the effect of intraoperative EHR downtime exposure on important patient-centered outcomes. The operating room is a data-dense and rich environment that relies on many electronic applications for coordinated care delivery. A lapse in these systems caused by EHR downtime can influence perioperative care delivery. We hypothesized that intraoperative exposure to EHR system downtime would cause an increase in the length of postoperative stay, the duration of time spent in the operating room, and 30-day mortality rates.

MATERIALS AND METHODS

Study setting and design

This retrospective cohort study included patients age 18 years or older who underwent inpatient surgical procedures from January 01, 2011 through December 31, 2016 at the Mayo Clinic in Rochester, MN. The 7 core clinical applications in the EHR that support care delivery in the operating room were evaluated. Data regarding downtime characteristics and patient data were collected from institutional information technology incident and clinical databases. The Institutional Review Board at Mayo Clinic approved the study. A waiver of informed consent was granted. Patients who did not authorize their medical records for research use were excluded.

Study definitions

Downtime incidences of 7 clinical applications used in the operating room setting were evaluated: 1) an anesthesia information management system, 2) a picture archiving & communication system (PACS), 3) the CPOE system, 4) an application supporting clinical documentation, 5) an integrated clinical information viewer that retrieves data from multiple sources, 6) the surgical information recording system, and 7) the surgical coordination system. Downtime characteristics collected included date and time of occurrence, as well as nature of disruption, such as limited functionality or impairment, compared to complete inaccessibility or outage, and scheduled or unscheduled outage. Downtime incidents that were scheduled, carried out for maintenance, or less than 60 minutes in duration were excluded. Over the study period, 176 downtime incidents meeting inclusion criteria were identified.

The principal exposure under investigation was intraoperative occurrence of unplanned downtime of at least 60 minutes in duration compared to no exposure. For patients with multiple eligible surgeries, only the first procedure was considered to maintain uniqueness. In the context of exclusion criteria (Figure 1), unscheduled downtime of any of the preceding 7 clinical applications for 60 minutes’ duration or greater were included for this analysis. The data sets from these institutional databases were merged using patient identifiers and timestamps for the purpose of this analysis.

Figure 1.

Study flow diagram for included downtime incidents from January 01, 2011 through December 31, 2016.

Study cohort

Sample size was based on available data. All patients age 18 years or older who underwent surgical procedures lasting at least 60 minutes in duration with an inpatient stay exceeding 24 hours within the study period were eligible for inclusion. The following exclusion criteria were applied: 1) patients without a valid research authorization, 2) outpatient surgeries, 3) procedures performed outside of the standard inpatient operating rooms at the Rochester campus of the Mayo Clinic, 4) procedures where patients did not undergo general anesthesia, and 5) an American Society of Anesthesiologists (ASA) physical status classification of unknown or maximum score of 6 (declared brain-dead patient whose organs are being removed for donor purposes). Demographic data collected included age, sex, and race/ethnicity. Clinical data collected included ASA physical status classification, surgical specialty, time spent in the operating room, postoperative length of stay, and 30-day mortality.

Outcomes

The primary outcome was postoperative hospital length of stay. Secondary outcomes were duration of time spent in the operating room and 30-day mortality.

Data analysis

Information on patient age, sex, race/ethnicity, ASA physical status classification, day of surgery, time of surgery, and surgical specialty was tabulated. After review of outliers, no outlier modifications were performed. Downtime-exposed patients were matched to unexposed patients by day of the week and time of the day (morning, afternoon, and overnight). This approach controlled for temporal variations in personnel, physical resources available in the operating room, and EHR user demand. As potentially confounding variables, matched patients were also paired by ASA physical status classification, surgical specialty, and emergency or non-emergency status. The Greedy matching algorithm method was used for ASA physical status classification (±1), day of week (±1), weekend day indicator (exact), and surgical specialty (exact). Emergent surgery and shift were also included in the matching criteria, although exact matching on these variables was not required. When exact matches were not possible, the absolute difference was minimized.¹² Multivariable regression with generalized estimating equations to account for the matched study design was used. Matching variables were included in the model as covariates to adjust for residual confounders. To test for a trend in the effect of downtime in those experiencing an outage, duration of downtime was included in the multivariable models as a linear predictor of outcome. Estimates per 1 hour of downtime were reported. Statistical analysis was performed using JMP Pro (SAS, Cary, NC) and SAS software (SAS, Cary, NC). A p-value less than 0.05 was considered statistically significant. All confidence intervals (CIs) were reported at the 95% level.

RESULTS

A total of 2047 downtime incidents were identified over the 6-year study period involving the 7 EHR applications being evaluated. And a total of 1854 events with a median of 1 minute and interquartile range (IQR) of 0–3 minutes were excluded. Another 176 incidents met inclusion criteria (Figure 1). The median duration for these episodes was 152 minutes (IQR 86–269). The majority were impairments rather than complete outages and typically occurred on weekdays in the morning hours from 7:00 am to noon. The integrated clinical information viewer, picture archiving and communication, and CPOE systems were the 3 most commonly affected applications, in order of decreasing frequency (Table 1).

Table 1.

Characteristics of system downtimes from January 01, 2011 through December 31, 2016 (n = 176)

Characteristic	Count (IQR or %)
Duration in minutes, Median (IQR)	152 (86 to 269)
Impairment (%)	145 (82)
Outage (%)	31 (18)
Time of occurrence (%)
Morning (0701 to 1200)	90 (51)
Afternoon (1201 to 1700)	44 (25)
Overnight (1701 to 0700)	42 (24)
Day of week (%)
Sunday	8 (4.5)
Monday	38 (21.6)
Tuesday	36 (20.4)
Wednesday	36 (20.4)
Thursday	29 (16.5)
Friday	18 (10.2)
Saturday	11 (6.3)
Applications affected (%)
AIMS	9 (5.1)
Clinical documentation	9 (5.1)
CPOE	29 (16.5)
EHR Integrated viewer	83 (47.1)
PACS	41 (23.3)
Surgical information recording	14 (7.9)
Surgical listing	17 (9.6)

Abbreviations: AIMS, Anesthesia information management system; CPOE, Computerized provider order entry; PACS, Picture archiving and communication system.

In total, 87 024 patients were eligible for inclusion in the study. Of 4115 downtime-exposed patients, 4103 patients were successfully matched with unexposed patients by ASA physical status classification, emergency status, day of procedure, time of procedure, and surgical specialty. The demographics of these 2 groups were similar. For example, median age for both groups was 61 years with comparable sex and race/ethnicity distributions (Table 2). Most of the study population was comprised of patients with ASA physical status classification of 2 and 3 (maximum score of 6). General surgery, orthopedic surgery, and cardiac surgery were the most common surgical specialties, in order of decreasing frequency. There were no missing patient data.

Table 2.

Baseline characteristics of study patients

Characteristic	EHR Downtime-Exposed ( n  = 4103)	Not Exposed ( n  = 4103)	Percent of total (for matched variables)
Age in years, Median (IQR)	61 (49 to 71)	61 (50 to 71)
Male sex, n (%)	2215 (53.9%)	2155 (52.4%)
Race/ethnicity
White, n (%)	3657 (89.1%)	3723 (90.7%)
Other, n (%)	433 (10.6%)	370 (9.1%)
Missing, n (%)	13 (0.3%)	10 (0.2%)
ASA physical status class
1 (I), n	205	205	5.0%
2 (II), n	1870	1870	45.6%
3 (III), n	1826	1826	44.5%
4 (IV), n	197	197	4.8%
5 (V), n	5	5	0.1%
Emergency procedures, n	79	79	1.9%
Surgical specialty, n
Breast	45	45	1.1%
Cardiac	638	638	15.6%
Electrophysiology	8	8	0.2%
ENT	162	162	3.9%
General	1014	1014	24.7%
Gynecologic	105	105	2.6%
Neurologic	524	524	12.8%
Obstetrics	5	5	0.1%
Oral and Maxillofacial Surgery	36	36	0.9%
Orthopedics	719	719	17.5%
Pain	3	3	0.1%
Plastic Surgery	40	40	1.0%
Radiology	9	9	0.2%
Thoracic Surgery	209	209	5.1%
Transplant	88	88	2.1%
Urology	252	252	6.1%
Vascular Surgery	101	101	2.5%
Unknown	145	145	3.5%
Time of day of procedure start
Morning (7 am to noon)	3284	3284	80.0%
Afternoon (Noon to 5 pm)	766	766	18.7%
Overnight (5 pm to 7 am)	53	53	1.3%
Day of week
Sunday	15	15	0.4%
Monday	954	954	23.3%
Tuesday	851	851	20.7%
Wednesday	1030	1030	25.1%
Thursday	787	787	19.2%
Friday	444	444	10.8%
Saturday	22	22	0.5%

Abbreviations: ASA, American Society of Anesthesiologists; ENT, ear, nose, and throat; IQR, interquartile range

Downtime-exposed patients had operating room duration 1.1 times longer (CI, 1.08-1.12, p < .001) and postoperative length of stay 1.04 times longer (CI, 1.01-1.08, p = .007) compared to unexposed patients. The 30-day mortality rates were similar between the 2 groups by odds ratio of 1.26 (CI, 0.81-1.98, p > .05). In the trend analysis, there was no association between duration of downtime and evaluated outcomes, postoperative length of stay, and 30-day mortality (Table 3). A comparison of outcomes between these 2 groups showed downtime-exposed patients had a statistically significant increase in duration of operating room time of 1.10 times longer (CI, 1.08-1.12, p < .001) and postoperative length of stay 1.04 times longer (CI, 1.01-1.08, p = .007). The difference in 30-day mortality rates in exposed versus unexposed patients was not statistically different by odds ratio of 1.26 (CI, 0.81-1.98, p = .307). Tests for a trend in duration of downtime, postoperative length of stay, and 30-day mortality resulted in no evidence of trend effect (Table 4).

Table 3.

Patient outcomes

Characteristic	Estimate, Odds ratio	p-value
Duration of operating room time (95% CI)	1.10 (1.08–1.12)	< .001
Postoperative LOS (95% CI)	1.04 (1.01–1.08)	.007
30-day mortality (95% CI)	1.26 (0.81–1.98)	.307

Abbreviation: CI, confidence interval.

Note: Estimates are from multivariable, linear, or logistic regression as appropriate. Estimates are odds ratios for exposed compared to unexposed for categorical outcomes. For exposed compared to unexposed, odds ratios are multiplicatively increased for continuous outcomes.

Table 4.

Trend analysis evaluating effect of downtime duration on outcomes

Outcome	Estimate per 1 hour ofdowntime (95% CI)	p-value
Postoperative length of stay	−0.05 (−0.2–0.1)	.606
30-day mortality	0.99 (0.98–1.01)	.376

Abbreviation: CI, confidence interval.

Note: Estimates are parameter estimates for continuous outcomes and odds ratios for categorical outcomes.

DISCUSSION

The objective of this study was to evaluate the effect of intraoperative EHR downtime exposure on important patient-centered outcomes. We hypothesized intraoperative exposure to EHR system downtime would cause an increase in the length of postoperative stay, the duration of time spent in the operating room, and 30-day mortality rates. The results of this retrospective cohort study demonstrate EHR systems are susceptible to downtime of varying duration.

The few studies evaluating system downtimes show an adverse effect. In 1 survey, 57% of clinical nurses reported downtimes as being “disruptive” or “very disruptive” to workflow.¹³ In another statewide survey, pharmacy system downtimes in 78 hospitals led to medication errors.¹⁴ A study evaluating downtime of pathology testing and reporting systems demonstrated a consequent delay in clinician review of clinical results.¹⁵ Downtimes have been identified as a cause of a significant proportion of “events likely to impact care delivery” in a review of safety events related to England’s national program for information technology.¹⁶ Patient safety events resulting from system downtimes in the US have also been identified.¹⁷ In a study by the US Department of Health and Human Services, 24% of institutions surveyed reported resultant delays in patient care.¹⁸ It remains unclear how the extent and/or cause of the impact of downtime on clinical workflow and, consequently, the ability to deliver care may translate into poor patient outcomes. Delivery of health care currently depends on health information technology, which further reinforces the importance of these questions.

Although there is no apparent impact on mortality, downtime exposure may have a negative effect on other clinical outcome measures (eg, possibly prolonging the postoperative length of hospital stay and lengthening the duration of time spent in the operating room). A trend analysis did not reveal any association of downtime duration with postoperative length of stay and 30-day mortality rates. Results from the primary analysis indicate a possible association between exposure to downtime and perioperative outcomes, which is consistent with the findings of prior studies that evaluated system downtimes, such as adverse effect on clinical workflow and even patient harm.^8,14,15,19 However, it is notable that no trend effect was found in the secondary analysis. Only unscheduled downtime events greater than 60 minutes in duration were included in this study. Scheduled downtimes were excluded since there is a smooth transition to downtime procedures and they are planned around times of decreased user demand and/or fewer procedures. However, this raises the possibility that a greater proportion of emergent procedures may be exposed to and affected by downtimes.

The primary findings from this study can be explained by several possible mechanisms. Surgical patients receive care in the operating room, which requires a high level of coordination and teamwork among providers fulfilling different clinical roles with the aid of a variety of EHR applications. The anesthesia information management system is an integral tool for not only clinicians in the operating room but also for clinicians in the post-anesthesia care unitwho are engaged in the immediate care of patients after surgery. Similarly, the clinical documentation entry and viewer applications are used to generate notes, which are important mechanisms for the transfer of information, such as intraoperative complications, as patients move through the physical surgical workflow. It is possible that unavailability of these systems could cause incomplete transfer of critical information between the operating room, post-anesthesia care unit, and hospital floors leading to a delay in the appropriate diagnosis and/or treatment of what would otherwise be minor postoperative complications. Another key system where downtime can have immediate effects on a surgical procedure is the PACS, which is used for preoperative and intraoperative surgical planning. Inability to access this imaging system could result in a prolonged surgical approach and/or inadvertent complication, such as bleeding from transecting an anatomically aberrant blood vessel, which could otherwise have been known and avoided. As our results suggest, EHR applications contain an abundance of vital clinical information necessary for both the intraoperative and postoperative care of surgical patients, and it is possible that unavailability of these systems could have adverse consequences.

The possible consequences of application downtime episodes can be dichotomized by both “intermediate” and “downstream” effects on health care delivery. The immediate impact on care delivery can be represented by a prolongation of intraoperative duration, such as lack of PACS imaging, prolonging the time needed to resect a mass abutting critical organs and/or blood vessels. Given the expense of operating room time,²⁰ the increase we observed of 10% longer duration of time spent in the operating room has potential cost implications, as well as other clinical implications. For example, prolonged operative time in isolation has already been correlated with complications.²¹ The downstream effects of EHR downtime, represented by longer postoperative length of stay, can represent the sum of the immediate and subsequent effects, such as gaps in handoff communication and/or documentation. Our finding that EHR downtimes can extend hospital course by about 4% has both cost and patient-centered implications, even if overall mortality is unaffected.

These findings support the need to maintain updated contingency plans in the event of a downtime, as advocated at national levels—such as the SAFER Guides (Safety Assurance Factors for Electronic Health Record Resilience) by the Office of the National Coordinator for Health Information Technology—and institutional levels, where work-area–specific downtime preparedness plans are implemented.^22,23 Preparedness plans used at our institution include the CLEAR system: Check and communicate the problem, Locate the system downtime plan, Establish alternative patient care processes, Activate downtime plan, and Recover by entering data once downtime is over. The extent to which these backup plans are effective in minimizing consequences of EHR system downtime is unclear and merits further investigation. The evaluated 7 core clinical applications in the EHR that supported care delivery in the operating room over the course of this study existed as components of an integrated clinical system (core EMR). In other words, deconvolution of the effect of the individual applications, as well as important interpretation of subsequent analysis for impact, was omitted from this study due to anticipated complexity of these results. From a practical perspective, these core applications function as a single EMR system in the OR environment. Thus, the specific impact of these 7 applications on workflow processes, as well as how downtime is mitigated through routine clinical contingency planning in this setting, remains unknown.

In the secondary analysis, we tested for a trend and found duration of exposure to downtime had no effect on the postoperative length of stay or 30-day mortality. A trend effect in duration of operating room time was not assessed, given that the downtime event could have started prior to the beginning of a procedure or could have ended after a procedure had been completed. One possible explanation is that clinical care may be hampered at any point during a system downtime as a random occurrence regardless of duration of the event. The lack of a trend effect can also be explained by the exposure variable included, as well as equal weight to the 7 EHR systems used in the operating room, when some systems were more susceptible to downtime than others (Table 1). In other words, systems prone to relatively longer downtimes may not affect outcomes. Another explanation is that downtime events do not have true impact on the clinical outcomes measured, as limitations of this study with respect to the results of the primary analysis could also influence a lack of trend effect.

The findings of this study indicate a need for future research that evaluates the impact of scheduled compared to unscheduled downtimes including specific systems and varying durations in different health care settings. Further questions are raised such as the effectiveness of backup plans for negotiating system downtimes. This study evaluated salient perioperative outcomes. However, there are other measures relevant to the operating room setting such as delays in case start time in the event of a system downtime. There is also a need to perform similar investigations in differing health care settings.

Limitations

The study has several important limitations. 1) This is a single-center retrospective study. 2) Our definition of the exposure variable places different EHR systems at an equitable level, whereas some systems may be more critical for care delivery than others. 3) The results are subject to both known and unknown confounders. While we matched for multiple confounding variables, such as surgical specialty, we were unable to match on specific procedure codes due to the number of combinations. This raises the possibility of an unequal case mix between the exposed and unexposed groups, and thus possible selection bias. 4) As our aim was to evaluate intraoperative duration as a possible outcome variable, we could not control for the possibility that longer procedures may be more likely to overlap with an EHR downtime event, which potentially confounds these associations. 5) We did not use year of surgery as a matching variable given this would have prevented a good match for some procedures over the study span of 6 years. As there have been clinically significant advances in some surgical procedures during the study period, the date procedures were performed introduces another possible confounder. 6) Our study was not designed to analyze immediate peri-surgical outcomes, such as downstream workflow and care activities, which further address important challenges regarding the full extent of the impact of intraoperative EHR downtime. 7) Any potential relationship between resilience and intraoperative EHR downtime was not analyzed. In this context, it is not possible to generalize the results of this study at our institution to the potential impact of resilience and specific contingency planning to other hospitals.

One of the strengths of this study is that our institution maintains robust, longitudinal databases of application downtime incidents, as well as surgical patients and their outcomes, which allowed accurate identification of exposed patients. We identified a large sample of patients exposed to downtime events and also had great success in matching nearly every case to a control set of downtime-unexposed patients. Our sample was thus highly representative of our overall patient population and enhances generalizability, despite its single-center study design.

CONCLUSION

In this retrospective matched cohort study of the effects of intraoperative EHR downtime events among surgical patients, EHR downtimes had no influence on hospital mortality. Downtime-exposed patients had statistically significant longer intraoperative courses and postoperative hospital length of stay. These results have face validity and potential implications for both cost and patient care. These results emphasize not only the need for rigorous downtime preparedness plans, but also for further research regarding the implementation and effectiveness of these efforts. This is especially important given the increasing reliance of care delivery on health information technology and the growing vulnerability of this infrastructure—in part the result of increasing system complexity and external cybersecurity threats.

FUNDING

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

AUTHOR CONTRIBUTIONS

All authors (AMH, RS, BWP, and VH) contributed to the design of this study, data collection, data analysis, and manuscript preparation.

CONFLICT OF INTEREST STATEMENT

None declared.