Use of Electronic Health Records and Quality of Ambulatory Healthcare

Background This study aimed to measure the association between electronic health record (EHR) use and quality measures in ambulatory healthcare. Methodology A quantitative, retrospective, cross-sectional design was used by examining secondary data from the 2015-2016 National Ambulatory Medical Care Survey. The relationship between EHR use and seven quality measures was examined using the Donabedian model as a framework. Quality measures included (a) diabetes measures, (b) obesity measures, (c) blood pressure screening, (d) depression screening, and (e) breast cancer screening. A total of 37,290 office visits were included, representing 817 million national office visits. For each of the quality measures, we determined the (a) associations using unadjusted and adjusted regression models based on subsets of the sample that met the inclusion criteria for quality measures; and (b) the changes in the area under the curve (AUC). Results Approximately 75% of office visits fulfilled all EHR use. Positive associations were found between EHR use and better quality for the following three out of seven measures: higher odds of screening for obesity (odds ratio (OR) = 2.2; p = <0.0001), blood pressure (OR = 2.5; p = <0.0001), and breast cancer (OR = 1.8; p = 0.0166). Receiver operating curve results showed the highest gain in the AUC for process-grouped measures. Hence, it was considered to be a strong predictor for all quality measures. Conclusions Evidence showed improvement in some quality measures (screening for obesity, blood pressure, and breast cancer). Common and standardized health processes were more likely to be completed and recorded than others. Future policies concerning health information technology can shift the focus from improving EHR use to enhancing patient and quality outcomes. Further research is needed to identify circumstances where quality is improved.


Introduction
In the United States, there has been an increasing interest in health information technology (HIT). For example, electronic health records (EHRs) were mandated after 2014 as part of larger healthcare reform, with the goal of improving multiple aspects of care, including quality, safety, and cost [1,2]. EHR adoption increased rapidly between 2008 and 2015 from 42% to 89% among office-based physicians [3].
Healthcare providers are now required to report quality metrics called Clinical Quality Measures (CQMs) to the Centers for Medicare and Medicaid Services (CMS) to receive incentives and avoid penalties under the EHR Incentive Programs [4,5]. CQMs are tools that help measure and track the quality of healthcare service delivery [1].
Currently, billions of dollars are spent on adopting EHR systems and fulfilling reporting requirements for quality measures [6,7]. Yet, the literature presents mixed results regarding the relationship between EHR use and quality measures. Multiple studies have indicated that EHRs have clear advantages over paper records, such as accessibility, availability, and portability [8][9][10][11]. A systematic review concluded that overall EHR use could improve quality indicators due to better adherence to medical guidelines and reduced inappropriate care [12]. Kern et al. (2013) reported a positive correlation between physicians' use of EHRs and ambulatory care quality in a community-based setting. They found that four out of nine individual measures showed significant improvements and that EHR use generally was positively correlated with better quality of care compared to paper records [13].
However, EHRs do not consistently deliver positive impacts on quality indicators [14,15]. A comprehensive scoping review presented mixed results after examining the relationship between HIT and quality of care in the federal healthcare system [16]. Furthermore, clinical decision support systems seem to improve physicians' adherence to guidelines, though evidence does not indicate a statistically significant association [16]. Another systematic review assessed the impact of EHR use on structure, process, and outcomes by applying the Donabedian model in primary care [14]. The study reported that EHR use positively impacted structure and process compared to paper records [14] and was associated with improved quality measures, such as rates of screening measures for blood pressure, depression, and breast cancer [13,17,18]. EHR use has been linked with multiple beneficial diabetes-related outcomes when compared to paper records [19]. For instance, EHR use was linked to higher rates of HbA1c testing among patients with diabetes [13]. However, other studies did not find that consistent use of EHRs improved the quality of diabetes care [20].
Mixed results have been reported for obesity screening and prevention measures. Tracking obesity measures improved after the adoption of EHRs in a small practice in north Texas [21]. Likewise, there are higher odds of obesity screening (body mass index (BMI) documentation) with EHR use [22]; however, another study indicated a lack of BMI documentation in EHR records [23]. Romano and Stafford found that using EHRs was linked to higher rates of obesity counseling [24]; however, other studies found insignificant differences in exercise and dieting rates among high-risk adults [22,25].
Given these conflicting findings, our goal was to examine the relationship between EHR use and seven quality measures in ambulatory healthcare using elements of the Donabedian model. The Donabedian model has been used elsewhere to assess the quality of care in health services [26,27]. This study only examined structure and process measures but not outcomes [26,27].

Study design and sample
This study used a quantitative, retrospective, cross-sectional design. Secondary data were obtained from the 2015-2016 National Ambulatory Medical Care Survey (NAMCS) [28]. NAMCS is a national probability sample survey based on a sample of visits to non-federally employed office-based physicians in the United States [28]. This study received Institutional Review Board exemption due to de-identified medical records as the primary data source.
The response rate for physicians having at least one sampled visit was 30% for 2015 and 39% for 2016 [28]. Researchers restricted the analytic sample using the following criteria: (a) fulfilling the inclusion criteria of at least one of the seven quality measures, and (b) patients aged 12 years or older. Visit records with missing EHR use status were excluded (46 visits). A total of 37,290 office visits met the inclusion criteria. However, each quality measure had a unique population with a defined age parameter (see Table 1).

Quality measures
Proxy quality measures were inspired by literature and the CQMs used in the CMS Incentive Programs [29].
Six screening measures and one education/counseling measure focused on the following five different conditions: (a) diabetes ((i) eye exam screening, (ii) HbA1c testing); (b) obesity ((i) BMI measurement, (ii) education); (c) blood pressure screening; (d) depression screening; and (e) breast cancer screening (see Table  1). All quality measures were dichotomized to indicate whether appropriate care was provided.
All quality measures were tested using a single variable except for obesity education and blood pressure screening. Availability of both systolic and diastolic readings was used for blood pressure screening, and delivery of at least one of the three education/counseling services (diet/nutrition, exercise, and weight reduction counseling) was used for obesity education. Obesity screening was measured using BMI regardless of the recorded measurement.

Electronic health records
The level of EHR use was recorded as Yes for all EHR use and No for partial/no EHR use.
To fulfill the statistically minimum number of subjects in each cell, some answer categories were collapsed, such as "physician specialty," "patient referred to visit," and "payment for visit." Missing data were grouped with "other" categories.

Statistical analysis
Descriptive statistics (chi-square) and multivariate logistic regressions were used to examine the association between EHR use and quality measures. The Donabedian model was used as a conceptual framework in the analysis to explore the organizational structure and process components related to EHR use.
Researchers conducted two different analytical approaches. First, unadjusted and adjusted regression models were for each quality measure. Covariates included structure, process, and demographics. A p-value of less than 0.05 was used to determine statistical significance. Sample sizes for regression analysis ranged from 2,087 to 37,290 office visits depending on the quality measure's inclusion criteria.
Second, researchers used a receiver operating curve (ROC) approach by examining changes in the area under the curve (AUC). Covariates were grouped into three categories, namely, structure, process, and demographics. Starting with EHR use, consecutively adding elements of the Donabedian model (structure and process), then demographics for a total of four models for each quality measure [30]. All analyses took into consideration the complex survey design and were performed using SAS software version 9.4.

Results
This study included 37,290 office visits, representing an annual estimate of 817.6 million national office visits to ambulatory care physician offices. Office visits were stratified by EHR use status; all EHRs were used in 75% of visits, and partial/no EHRs in 25% of visits. Descriptive statistics showed that most patients were white or non-Hispanic/Latino (see Table 2).  Comparing visits in both EHR use groups, researchers found that the northeast region had the highest rate of visits with partial/no EHR use (31%), while the south region had the highest rate of visits with all EHR use (38%). Psychiatry had higher visit rates with partial/no EHR use (13%) compared to visits with all EHR use (4%). More than one-third of patients (36%) spent more time with their physician when partial/no EHR was used compared to all EHR use (24%) (see Table 2). Table 3 and Table 4 present unadjusted and adjusted regression models. Researchers found statistically significant positive associations between EHR use and three quality measures, namely, higher odds of obesity screening, breast cancer screening, and blood pressure screening with all EHR use.    We controlled for (1) system replacement plans as a structure measure and (2) survey year as a process measure in the logistic regression, but OR and p-values are not presented in the table.

EHR use Yes, all EHR None or partial EHR use P-value
NS: Non-sufficient sample size for the valid odds ratio.
NAMCS: National Ambulatory Medical Care Survey; EHR: electronic health record; OR: odds ratio

Diabetes measures
An adjusted regression model did not find a significant association between EHR use and eye screening among diabetes patients, though the results were borderline significant (p = 0.0514) in the unadjusted model. Patients with diabetes were more likely to get eye screening if they were seen by an ophthalmologist (odds ratio (OR) = 169.95) or referred for a visit (OR = 3.7), but they were less likely to get screened if they were 12-17 years old, visited the facility more than six times, or if payment for the visit was unknown or patient pay.
There was no significant association between EHR use and diabetes HbA1c testing. Diabetic patients were more likely to get tested for HbA1c if they saw their primary care provider (OR = 5.1) or spent more than 30 minutes with the provider (OR = 1.7), but they were less likely to get tested for HbA1c if they were 12-17 years old ( Table 3).

Obesity screening and education
Office visits with all EHR use were more likely to be screened for obesity when compared to visits with partial/no EHR use in both adjusted (OR = 2.2) and unadjusted (OR = 2.6), regression models.
Patients were more likely to be screened for obesity if they saw their primary care provider (OR = 2.8), spent more than 30 minutes with the provider (OR = 1.4), or had no previous visits (OR = 1.4). They were less likely to be screened for obesity if they were younger than 50 years of age or 65 years and over, saw specialists (OBGYN, ophthalmologist, or psychiatrist), were established patients at the facility, or had more than six visits.
There was no significant association between obesity education among overweight and obese patients in both adjusted and unadjusted models. Patients were more likely to receive obesity education if they were non-Hispanic or Latino (OR = 1.7), saw their primary care provider (OR = 1.7), and were established patients at the facility (OR = 1.4). They were less likely to receive obesity education if they were less than 50 years of age ( Table 3).

Breast cancer screening
Females visiting offices with all EHR use were more likely to receive mammograms when compared to visits with partial/no EHR use for both adjusted (OR = 1.8) and unadjusted (OR = 2.2) regression models. Furthermore, females were more likely to receive mammograms if they saw their primary care provider (OR = 3.3), were seen by an OBGYN physician (OR = 12.0), and had no previous visits to the facility (OR = 3.0); however, they were less likely to receive mammograms if they were established patients at the facility ( Table 4).

Depression screening
There was no significant association between EHR use and depression screening. Patients were more likely to be screened for depression if seen by their primary care provider (OR = 5.3) or psychiatrist (OR = 7.9); however, they were less likely to be screened for depression if they were seen in the west or were established patients ( Table 4).

Blood pressure screening
Office visits with all EHR use had higher odds of blood pressure screening compared to visits with partial/no EHR in both adjusted (OR = 2.5) and unadjusted (OR = 2.8) regression models. Patients were more likely to get their blood pressure checked if they were non-white (OR = 1.6), saw their primary care provider (OR = 10.4) or an OBGYN physician (OR = 7.6), spent more than 30 minutes with the provider (OR = 1.6), and used Medicare/Medicaid for the visit (OR = 1.2); however, they were less likely to get their blood pressure checked if they were less than 50 years of age or seen by specialists (ophthalmologist or psychiatrist) ( Table 4). Table 5 shows the ROC analysis. The demographic group was not or a weak predictor across all quality measures with a gain in AUC ranging from 0.000 to 0.004. EHR use was a weak predictor for most quality measures except for obesity and blood pressure screening. Likewise, the structure group was a weak predictor except for diabetic HbA1c and depression screening. The process group was a strong predictor for all quality measures with a gain in AUC ranging from 0.117 to 0.280.  *: The final result represents the full models shown in Table 3 and Table 4.
NAMCS: National Ambulatory Medical Care Survey; ROC: receiver operator curve; AUC: area under the curve; EHR: electronic health record

Discussion
This study used NAMCS 2015-2016 survey data to examine associations between EHR use and seven quality measures. We found significant positive associations between EHR use and better quality for three quality measures. The adjusted and unadjusted regression models' results were reasonably similar suggesting that adding additional structure, process, and demographic factors did not have a substantial effect on quality measures. However, the ROC analysis results showed a noteworthy increase in AUC after adding process measures.
The findings of this research are consistent with prior studies which suggested that EHR use was not always positively associated with improvements in quality measures [15,17]. This research updated and  [15,17,25]. They compared EHR use to paper records, while this study compared it to partial/no EHR use. Given current EHR use requirements, there just were not enough offices with paper use only.
Further, we based our analysis on a theoretical framework by considering structure and process as the main components. Structure and process are essential because they can help predict outcomes of care delivered in physician offices. More details on the Donabedian models will be discussed in the following sections. Each quality measure will be discussed separately.

Diabetes measures
In this study, there were no correlations between EHR use and both diabetes measures (HbA1c and eye exam). It is noteworthy that the diabetes eye exam was borderline significant (negative association) in the unadjusted model. Contradicting our results, researchers found that EHR use improved diabetic individual measures (HbA1c and retinal exam) and composite compared to paper records [19]. Moreover, the rates of HbA1c testing were higher with EHR use in an ambulatory care setting, but diabetic eye exams did not show improvements [13]. The latter finding is consistent with our results. Similarly, when diabetic patients were followed for three years, EHRs did not ensure improving quality [20].
Several reasons can explain these contradictory findings. Cebul et al. (2011) used older data from NAMCS (2009-2010) to calculate composites of diabetic care [19]. Kern et al. (2013) compared EHRs to paper records in the Hudson Valley of New York, while this study compared all EHR use to partial/no use on a national level from more recent NAMCS data [13].

Obesity screening and education
Patients were more likely to receive obesity screening with all EHR use among visits to physician offices in both adjusted and unadjusted models. A study found that using EHRs improved BMI documentation, but no improvement in developing follow-up plans among overweight and obese patients [22]. Findings from that study confirmed our results. However, other researchers found that BMI documentation and obesity diagnosis were not always provided in EHR systems, which indicates either lack of documentation or an absence of screening [15,23]. Although this study found an association between EHR use and obesity screening, only 41% of visits had a documented BMI, implying that obesity screening is not always performed or recorded. Being diagnosed with obesity is a strong predictor of formulating a weight management follow-up plan. Therefore, it is crucial to improve obesity screening to prevent and control the obesity epidemic. Romano and Stafford (2011) found that EHR use was related to improved diet and exercise counseling [24]. Their findings contradict our results; this might be due to the comparison categories of EHRs. They compared EHR use with paper records, while this study compared it with partial/no EHR use.

Breast cancer screening
Female patients were more likely to receive mammograms with all EHR use among visits to physician offices in both adjusted and unadjusted models. This is consistent with another study that found higher odds of breast cancer screening being associated with EHR use in a community-based setting [13]. In contrast, another study found that EHR use was not associated with improving women's health, including breast cancer screening [15]. This contradiction could be attributed to the way Poon et al. (2010) examined outcomes as groups under women's health or cancer prevention rather than breast cancer outcomes individually.

Depression screening
Contradicting our results, Akincigil and Matthews (2017) found that physicians who fully adapted EHRs were more likely to screen for depression when compared to physicians using paper records, even though depression screening was generally low [18]. Our findings also indicated that only 3.6% of eligible patients received depression screenings, which is considered fairly low. However, in our study, no correlation was found between depression screening and EHR use in ambulatory care. This discrepancy might be attributed to the different comparison groups.

Blood pressure screening
Patients were more likely to have blood pressure checked with all EHR use among visits to physician offices in both adjusted and unadjusted models. On the contrary, others found no difference or lower odds of blood pressure checks with any EHR use [17,24,25]. Although all three studies used NAMCS, they examined older data [25]. One explanation might be that EHR systems have advanced over time and/or medical professionals are now more accustomed to using these systems efficiently. It might also be due to the comparison categories of EHR use.

Structure and process
Researchers believe incorporating the Donabedian model in the analysis is helpful as structure and process components play vital roles in facilities' capabilities and eventually can affect attributes and quality of care delivered. According to the Office of the National Coordinator for HIT, structure and process are commonly used quality measures in healthcare settings.
In our findings, no structure measures had a significant association with any quality measure except one negative association: physician offices in the western region had lower rates of depression screening. However, each individual process measure had a significant association with at least two quality measures, and, in some cases, with most quality measures, such as "primary care provider." Previous literature did not focus on structure and process as main components. While some studies presented structure and process in crosstab results, no studies showed regression results. Therefore, our findings are a new addition to the literature. However, a systematic review used the structure and process components as outcomes in primary care and concluded that EHRs positively impacted both components compared to paper records. This research used the structure and process measurements as covariates (independent variables), while other studies used them as dependent variables [14].
Despite limited to no associations between structure or process and quality measures, the ROC analysis showed some interesting results. Because process measures already showed some significant results in the regression analysis, they showed the highest gain in AUC for all quality measures. Structure components showed a reasonable gain in AUC, especially for depression screening, despite no regression analysis associations. In regression models, each structural and process measurement is tested independently, reflecting minor changes in quality measures representing insignificant results. However, these variations are magnified when variables are grouped together, explaining the ROC results. AUC is commonly used when a large number of variables are studied and can be grouped together.

Limitations
The cross-sectional design of the study prevents inferring causality between variables. Secondary data analysis limited the ability to collect desired variables. As we were not able to include all components of the Donabedian model, analysis was restricted to structure and process but not health outcomes. NAMCS did not provide data related to patient outcomes such as morbidity and mortality. We wanted to include physician office size for structure measures, but the variable was masked in the dataset. Office size can significantly define the capabilities of the facility, as well as EHR capabilities. NAMCS response rate is considered relatively low for both years, which can cause nonresponse bias.
In addition, we did not control for patient compliance which might affect the rates of screening and counseling. Our analysis of visit records indicated if tests or services were performed, but no information was found related to physician's orders that were not performed.
The selected quality measures only represent specific conditions and characteristics of care; therefore, it does not represent the overall care aspect. It is important to note that some associations may be caused by chance alone, and some associations are difficult to justify or explain. The perplexity and scarcity of these associations do not necessarily mean they are nonexistent. Furthermore, for a few quality measures, the small sample size may have prevented us from finding statistically significant associations.

Conclusions
This research found positive associations between EHR use and better quality for three out of seven quality measures (obesity screening, blood pressure screening, and breast cancer). EHR might improve some clinical measures but not consistently improve all quality measures. We are not refuting that EHRs can improve quality in healthcare facilities in the United States, but findings suggest that as efforts to improve EHR use and adoption expand, automatic improvement of quality should not be assumed as a result. Moreover, measuring clinical quality should be continuously studied due to rapid advancements and changes in EHR systems over time. Evidence show improvement in some quality measures. Common and standardized health processes were more likely to be completed and recorded than others.
Policymakers are highly encouraged to consider focusing on future EHR criteria and efforts to improve prevention, screening, and counseling rates to facilitate quality care for populations. HIT future policies can shift the focus from improving EHR use to enhancing patient and quality outcomes. Future research should examine all three components of the Donabedian model to translate clinical quality outcomes (quality measures) to patient quality outcomes. Additionally, research should also focus on validating these findings on current EHR use because EHRs are advancing rapidly. Further research is needed to identify circumstances where quality is improved. Clinical decision support systems should be incorporated in future research to find details about associations between EHR and clinical quality measures. Clinical outcome improvement does not necessarily translate to patient outcomes, such as health improvements.

Additional Information Disclosures
Human subjects: All authors have confirmed that this study did not involve human participants or tissue. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue.

Conflicts of interest:
In compliance with the ICMJE uniform disclosure form, all authors declare the following: Payment/services info: All authors have declared that no financial support was received from any organization for the submitted work. Financial relationships: All authors have declared that they have