Mammography Use among Black Women: The Role of Electronic Medical Records

Abstract

Background and Aims:

Accurately documenting mammography use is essential to assess quality of care for early breast cancer detection in underserved populations. Self-reports and medical record reports frequently result in different accounts of whether a mammogram was performed. We hypothesize that electronic medical records (EMRs) provide more accurate documentation of mammography use than paper records, as evidenced by the level of agreement between women's self-reported mammography use and mammography use documented in medical records.

Methods:

Black women aged 40–75 were surveyed in six primary care sites in Boston, Massachusetts (n = 411). Survey data assessed self-reported mammography prevalence within 2 years of study entry. Corresponding medical record data were collected at each site. Positive predictive value (PPV) of self-report and kappa statistics compared data agreement among sites with and without EMRs. Logistic regression estimated effects of site and patient characteristics on agreement between data sources.

Results:

Medical records estimated a lower prevalence of mammography use (58%) than self-report (76%). However, self-report and medical record estimates were more similar in sites with EMRs. PPV of self-report was 88% in sites with continuous access to EMRs and 61% at sites without EMRs. Kappa statistics indicated greater data agreement at sites with EMRs (0.72, 95% CI 0.56-0.88) than without EMRs (0.46, 95% CI 0.29-0.64). Adjusted for covariates, odds of data agreement were greatest in sites where EMRs were available during the entire study period (OR 4.31, 95% CI 1.67-11.13).

Conclusions:

Primary care sites with EMRs better document mammography use than those with paper records. Patient self-report of mammography screening is more accurate at sites with EMRs. Broader access to EMRs should be implemented to improve quality of documenting mammography use. At a minimum, quality improvement efforts should confirm the accuracy of paper records with supplemental data.

Introduction

Racial disparities persist in the presentation of advanced stage breast cancer and breast cancer mortality, with higher rates for black women compared with other groups.¹ Therefore, meeting Healthy People 2010 objectives for early breast cancer detection has been prioritized in primary care settings that serve women who are at risk for underscreening.^2
–4 To ensure high-quality care in early breast cancer detection, mammography use must be accurately documented in primary care settings. Electronic medical records (EMRs) have the potential to improve the quality of documentation for screening use in ambulatory settings.^5,6 The role of EMRs in improving the measurement of screening mammography use in underserved populations, however, requires further study.⁷

Clinical and public health studies have compared estimates of mammography use obtained from medical records with those obtained from women's self-report.⁸ Generally, medical record documentation is thought to provide more accurate accounting of mammography use than patient self-report; self-reported mammography use is thought to overestimate actual utilization.⁸ To our knowledge, however, differences in the methods for documenting mammography use in primary care settings have not been evaluated by whether electronic or paper records were used. Ideally, compared with paper records, EMRs improve documentation by providing easily found, legible, organized information that can interface with on-site or remote laboratory or radiology systems for efficient data retrieval.⁵ The availability of EMRs in primary care settings that care for underserved populations is increasing but remains limited.⁹ It is possible that differences in access to technology for documenting mammography use influence the ability to measure the quality of care provided to underserved groups.

We recently reported the results of a community-based participatory research intervention in six primary care settings that showed health center differences in longitudinal use of mammography among black women with high social and medical risks for underscreening (unpublished observations¹⁰). In the intervention study, we hypothesized that site-level infrastructure differences may have accounted for some of the variation in mammography use among black women at the different intervention sites. The present study evaluates the availability of an EMR as a potential contributor to these differences by affecting the accuracy of measuring mammography use. We hypothesize that data on screening mammography use that is self-reported at baseline by black women and the screening mammography use documented by medical record review at baseline will agree more frequently at sites that use EMRs as the medical record source than at sites that use paper records as a result of potentially more complete documentation provided by EMRs.

Materials and Methods

Study participants

The present study was conducted as a part of the Boston Racial and Ethnic Approaches to Community Health Breast and Cervical Cancer Coalition (REACH Coalition) Women's Health Demonstration Project (WHDP) intervention. The WHDP is a community-based participatory research intervention designed to decrease racial disparities in breast cancer mortality between black and white women by improving screening mammography use and follow-up of abnormal results among women of African descent. We previously reported the details of participant recruitment and cohort description (unpublished observations¹⁰). The WHDP mammography study recruited black women aged 40–75 who were at high risk for receiving fragmented mammography care because of a history of missed primary care or mammography appointments.

Enrollment of participants took place from 2002 through 2006. Black women were recruited among six participating primary care sites in Boston, Massachusetts (n = 483). Participants who had any known cancer or who were under surveillance for suspected breast cancer or breast problems at enrollment were not considered eligible for screening and were excluded from analysis (n = 39). Medical records (paper records) could not be located for 7 participants, who were thus excluded from analysis. We sought to evaluate documentation for women who received their primary care, rather than only other services (dental care, eye care), in the six enrollment sites. Thus, survey data were used to restrict the analysis to women who had regular providers at their site of enrollment. Twenty-six women who reported they did not have regular providers at their site of enrollment were excluded, leaving data from 411 women for analysis. The Institutional Review Boards overseeing all sites, Brigham and Women's Hospital, Boston Medical Center, Beth Israel Deaconess Medical Center, and Boston Children's Hospital, approved this study.

Study sites

The WHDP was conducted in six primary care sites: (1) an academic hospital clinic with on-site mammography and continuous availability of EMRs for mammography documentation during the study period, (2) a community health center licensed by an academic hospital that had continuous availability of EMRs for mammography documentation during the study period but did not have on-site mammography, (3) a free-standing community health center without on-site mammography that initially used a paper record system but transitioned to an EMR system early in the study enrollment period in 2003 (“early transition to EMR”), (4) a free-standing community health center with on-site mammography that initially used a paper record system but transitioned to an EMR system late in the study enrollment period between 2005 and 2006 (“late transition to EMR”), and (5) two free-standing community health centers without on-site mammography services and without EMRs in use for mammography documentation during the study period.

In this study, the sites that had EMRs systems in place or that transitioned to EMRs during the study each used systems with fully functional capacity for clinical documentation, including the ability to manage patients' demographics, problem lists, clinicians' documentation or notes, and the capacity to capture or store internal and external clinical documents, including mammography reports and laboratory tests.¹¹

Medical record abstraction procedures

Research assistants reviewed paper and electronic records at each site and were trained in the use of a standardized medical record abstraction tool to collect the date a mammogram was performed, the indication for the mammogram (screening or diagnostic test), the result of the mammogram (including the Breast Imaging Reporting and Data System (BI-RADS^®) category), and any diagnosis of cancer. Research assistants reviewed medical records at baseline in the year of study entry and abstracted data retrospectively on recent mammography use at baseline. Data were abstracted from sections of the paper or electronic medical record labeled “radiology,” “laboratory,” “pathology,” “outside reports,” or “health maintenance” and from providers notes. In sites that transitioned between paper and electronic technologies during the study period, research assistants abstracted all remaining paper charts as well as the electronic charts.

Measurement of recent mammography use at baseline

The present study analyzed the prevalence of recent mammography use at baseline, defined as a mammogram obtained in the past 2 years prior to study entry, to determine if Healthy People 2010 objectives for mammography use (70% prevalence of recent mammography use among women aged ≥40) were met at each site. At study entry, self-report of recent mammography use was obtained by a written survey administered in-person with a single item: Have you had a mammogram within the last 2 years? Medical record reports of recent mammography use were used to determine if participants had a mammogram within the past 2 years prior to study entry, as described above. Medical record data on mammography obtained within the past 3 years prior to baseline were also reviewed to assess for evidence of telescoping, which is the tendency to report having a mammogram more recently than the test was actually performed.

Covariates

Survey data also captured information on factors thought to influence the prevalence of mammography use estimated by self-report, including annual household income, insurance type, family history of breast cancer, experience of discrimination, women's satisfaction with communicating with their providers, and women's satisfaction with information on breast health provided by their clinics. Additionally, our recruitment procedures sought to enroll diverse women of African descent, including those who were not U.S. born. Thus, analyses were adjusted for nativity to capture the possibility of differences in measurement as a result of receiving care outside of the United States, which might not be captured in medical record reviews.

Statistical analysis

We compared self-report and medical record review estimates of recent mammography use at baseline. Medical record review was considered the gold standard of reporting. The prevalence, positive predictive value, and negative predictive value of self-report of having a mammogram in the 2 years prior to study entry, relative to medical record review, were computed for each site to compare values for sites with paper records or EMRs. Agreement between medical record documentation and self-report was calculated for each site as the percentage of positive and negative accounts of mammography use that matched between self-report and medical record review. Kappa statistics were calculated to determine the extent of concordance above chance agreement between medical record review and self-report for each site. Predictors of the odds of agreement between medical record review and self-report were estimated by logistic regression.¹² Predictors of agreement corrected for chance via the kappa statistic were estimated with generalized estimating equations.¹³ All statistical procedures were conducted in SAS 9.1 (Cary, NC).

Results

Participant and enrollment site characteristics

Table 1 shows participant characteristics by site. All sites were similar in terms of participants' annual household income, family history of breast cancer, and experience of discrimination. Statistically significant differences among sites are shown in Table 1. Specifically, study participants at the site with continuous EMRs and on-site mammography were older than women at other sites. The site with continuous EMR but off-site mammography had the highest immigrant population. The site with early transition to EMR had the highest proportion of patients with public insurance, either Medicare or Medicaid, and the site with late transition to EMRs had the highest proportion of patients who were uninsured. The sites with no EMR available during the study had the highest proportion of patients with supplemental private insurance.

Table 1.

Selected Participant Characteristics by Site of Enrollment a

	p value	Continuous availability of EMR, b on-site mammography	Continuous availability of EMR, off-site mammography	Early transition to EMR, off-site mammography	Late transition to EMR, on-site mammography	No EMR, off-site mammography
	p value	n = 96	n = 96	n = 37	n = 107	n = 75
Mean age, years (SD)	<0.001	54 (8)*	51 (9)	52 (8)	52 (9)	49 (7)
Insurance coverage	<0.01
Uninsured (free care)		13 (14)	25 (27)	8 (23)	33 (32.7)*	10 (14)
Public (Medicare/Medicaid)		61 (64)	47 (50)	23 (66)*	41 (40.6)	38 (54)
Private supplemental		21 (22)	22 (23)	4 (11)	27 (26.7)	23 (32)*
Nativity	<0.001
U.S. born		90 (94)*	38 (40)	30 (81)	57 (53)	60 (80)
Non-U.S. born		6 (6)	58 (60)*	7 (19)	50 (47)	15 (20)
Income	0.33
<$10,000		46 (52)	44 (49)	18 (53)	58 (59)	34 (51)
$10–25,000		26 (30)	29 (33)	12 (35)	24 (24)	28 (42)
>25,000		16 (18)	16 (18)	4 (12)	17 (17)	5 (7)
Family history of breast cancer	0.12	19 (20)	19 (20)	7 (19)	19 (18)	24 (33)
Experience of discrimination	0.07	7 (7)	10 (10)	0	4 (4)	9 (12)

Figures represent n (%) unless otherwise noted. Where data are missing due to item nonresponse, numbers do not sum to 100%. General linear model tests associations between categorical and continuous variables. Chi-square tests associations among proportions. Cohort only includes participants with a regular clinical provider at baseline.

EMR, electronic medical record; early transition, adopted EMR during first year of study; late transition, adopted EMR during last years of study.

Highest values for statistically significant comparisons.

All sites were rated uniformly highly in the percentage of women who were satisfied with getting medical care and breast information at their clinic (Table 2). However, the clinics with no EMRs during the study had the highest prevalence of women with concerns about provider communication (Table 2).

Table 2.

Perceived Provider and Clinic Characteristics by Site of Enrollment a

	p value	Continuous availability of EMR, b on-site mammography	Continuous availability of EMR, off-site mammography	Early transition to EMR, off-site mammography	Late transition to EMR, on-site mammography	No EMR, off-site mammography
	p value	n = 96	n = 96	n = 37	n = 107	n = 75
Any concerns communicating with your healthcare provider?	<0.01
Yes		16 (17)	13 (14)	8 (22)	7 (7)	20 (28)*
No		78 (83)	83 (86)	29 (78)	100 (93)*	52 (72)
I feel comfortable when I come to get my care here.	0.22
Agree		87 (94)	94 (99)	31 (100)	100 (97)	69 (96)
No opinion/disagree		6 (6)	1 (1)	0	3 (3)	3 (4)
I feel comfortable getting information about breast problems at this clinic.	0.75
Agree		84 (91)	88 (94)	27 (87)	92 (94)	66 (92)
No opinion/disagree		8 (9)	6 (6)	4 (13)	6 (6)	6 (8)

Figures represent n (%) unless otherwise noted. Where data are missing due to item nonresponse, numbers do not sum to 100%. Chi-square tests associations among proportion. Cohort only includes participants with a regular clinical provider at baseline.

EMR, electronic medical record; early transition, adopted EMR during first year of study; late transition, adopted EMR during last years of study.

Highest values among statistically significant comparisons.

Recent mammography use estimated by self-report and medical record review

Across all sites, the prevalence of recent mammography use depended on the measure used (Table 3). By patient self-report estimates, the prevalence of recent mammography use (had a mammogram in the past 2 years at baseline) was 76%; by medical record review estimates, it was 58%. The positive and negative predictive values of self-report across all sites were 74% and 94%, respectively. The agreement between self-report and medical record review was 79%, with a moderately high kappa statistic (0.54, 95% CI 0.46–0.62) (Table 3). Generally, a kappa statistic between 0.40 and 0.60 is thought to indicate moderate agreement, and a kappa statistic exceeding 0.60 is thought to indicate strong data agreement.¹⁴

Table 3.

Agreement between Self-Report and Medical Record Documentation of Mammography a

	Medical record documents mammogram in past 2 years
	Yes	No	Total
Self-reported mammogram in past 2 years
Yes	233	81	314
No	6	91	97
Total	239	172	411
Predictive value of self-report	PPVb = 74%	NPV = 94%
% agreement between self-report and medical record			79%
Kappac (95% CI)			0.54 (0.46–0.62)

Participants aged 40–75 with a regular provider at baseline.

PPV, positive predictive value; NPV, negative predictive value.

Kappa tests probability of agreement between medical record and self-report of mammography use in 2 years prior to study entry beyond chance agreement.

Site differences in recent mammography use estimated by paper records and EMRs

The prevalence of mammography use and the agreement between medical records and self-report differed by site (Table 4). When mammography use was measured by patient self-report, the Healthy People 2010 objective for recent mammography use (70%) was met at most sites (Table 4). The centers with early transition to EMR (59%) and without EMR (65%) had self-reported mammography use that was below the 2010 goal. However, when mammography use was measured through medical record review documentation, the Healthy People 2010 objective was clearly met at only one site. Only the hospital-licensed community health center with continuous use of EMRs and off-site mammography exceeded the 2010 goal (72%). The 2010 goal was nearly met by the academic hospital clinic with continuous use of EMRs and on-site mammography (69%). By medical record review, the sites without EMRs continuously available during the study period did not meet Healthy People 2010 objectives, either with (52%) or without (43%) on-site mammography. Statistically significant site differences in the prevalence of baseline mammography were seen by either measure.

Table 4.

Estimates of 2-Year Mammography Use by Self-Report and Medical Record Review: Availability of EMR a

	p value b	Continuous availability of EMR, c on-site mammography	Continuous availability of EMR, off-site mammography	Early transition to EMR, off-site mammography	Late transition to EMR, on-site mammography	No EMR, off-site mammography
	p value b	n = 96	n = 96	n = 37	n = 107	n = 75
Prevalence data
Mammography use in prior 2 years
Had mammography, Self-report	<0.01	77%	80%	59%	86%	65%
Had mammography, medical record review	<0.001	69%	72%	43%	52%	43%
Data agreement
Percent agreement between self-report and medical record review	<0.001	88%	90%	84%	64%	72%
PPV		86%	88%	73%	60%	61%
NPV		91%	95%	100%	93%	92%
Kappad (95% CI)		0.69 (0.53-0.85)	0.72 (0.56-0.88)	0.68 (0.46-0.90)	0.27 (0.13-0.40)	0.46 (0.29-0.64)

Cohort only includes participants with a regular clinical provider at baseline. Participants were enrolled between 2002 and 2006.

p value tests statistically significant differences between sites.

EMR, electronic medical record; early transition, during first year of study; late transition, during last years of study; PPV, positive predictive value; NPV, negative predictive value.

Kappa measures probability of agreement between medical record and self-report of mammography use in 2 years prior to study entry.

Discrepancies between self-report and medical record review by availability of EMR

The agreement between medical record review and self-report of mammography use differed by site as well (Table 4). Highest rates of agreement between self-report and medical record review were seen at the two sites with continuous availability of EMRs (88% and 90%), and lowest rates were seen at sites with late transition to EMRs and without EMRs (64% and 72%). Sites with continuous EMR availability also had the highest positive predictive values of self-report relative to medical record review (Table 4). The site with early transition to EMR had higher positive predictive value of self-report (73%) than either the site with late transition to EMR (60%) or those without EMRs (61%). The negative predictive value of self-report was high for all sites, ranging from 91% to 100% (Table 4).

Kappa statistics indicated high levels of agreement between self-reports of mammography use and medical record review at sites with continuous and early availability of EMRs (Table 4). The sites with late transition to EMR and no EMR available had modest levels of agreement between self-report and medical record review over chance.

Telescoping as source of disagreement between self-report and medical record review

There was evidence of telescoping, meaning that women may have remembered their mammography event as being more recent than recorded by medical record review. Of the 314 women who stated by self-report that they obtained mammography 2 years prior to study entry, we identified 16 women whose record of mammography use was found in the medical record within 3 years prior, rather than 2 years prior to study entry (5%). However, 65 women who reported having a mammogram within 2 years prior to study entry did not have a medical record of a mammogram even within 3 years of study entry (21%).

Predictors of self-report and medical record agreement: Adjustment for covariates

Site of enrollment was a significant predictor of agreement between self-report and medical record review (Table 5). Compared with sites without EMRs during the study period, sites with continuous EMRs had higher odds of medical record and self-report agreement after adjustment for covariates (Table 5). A trend toward higher odds of agreement in the site with early transition to EMR compared with those without EMRs was not statistically significant (OR 1.97, 95% CI 0.62-6.19). The comparison between the site with late transition to EMR and the centers without on-site mammography or EMRs was not statistically significant.

Table 5.

Predictors of Agreement between Chart Review and Self-Report of Mammography a

	n (%)	% agreement between self-report and medical record review	Odds of agreement adjusted for listed covariates OR (95% CI)	Kappa adjusted for covariates β (SE)
Patient factors
Age ≥50	213 (52)	80	1.05 (0.60-1.86)	−0.095 (0.09)
Income
<$10,000	200 (53)	78	0.47 (0.17-1.32)	0.026 (0.18)
$10,000–$25,000	119 (32)	80	0.70 (0.25-1.95)	0.132 (0.19)
>$25,000	58 (15)	81	Reference
Insurance coverage
Public insurance (Medicaid/Medicare)	210 (53)	80	1.50 (0.66-3.39)	0.153 (0.15)
Uninsured/free care	89 (22)	79	1.70 (0.66-4.39)	0.091 (0.13)
Private insurance	97 (25)	78	Reference
Nativity
Non-U.S. born	136 (33)	76	0.62 (0.32-1.22)	−0.222 (0.11)*
U.S. born	275 (67)	80	Reference
Family history of breast cancer	88 (22)	67	0.43 (0.23-0.82)*	−0.129 (0.10)
Experience of discrimination	30 (7)	67	0.44 (0.17-1.13)	−0.248 (0.16)
Provider factors
Concerns communicating with my provider	64 (16)	81	1.60 (0.72-3.55)	0.161 (0.10)
Clinic factors
I feel comfortable getting information about breast problems at this clinic	357 (92)	79	0.78 (0.27-2.28)	0.189 (0.15)
Site of enrollment
Continuous availability of EMRb, on-site mammography	96 (23)	88	3.57 (1.40-9.13)*	0.344 (0.13)*
Continuous availability of EMR, off-site mammography	96 (23)	90	4.31 (1.67-11.13)*	0.406 (0.14)*
Early transition to EMR, off-site mammography	37 (9)	84	1.97 (0.62-6.19)	0.264 (0.16)
Late transition to EMR, on-site mammography	107 (26)	64	0.88 (0.40-1.93)	−0.072 (0.13)
No EMR, off-site mammography	75 (18)	72	Reference

Models assess agreement between medical record review documentation and self-report of having a mammogram in the 2 years prior to study entry, adjusted for listed covariates. All patients had a regular provider at study entry. Some items do not sum to 100% due to missing data from item nonresponse.

EMR, electronic medical record; early transition, adopted EMR during first years of study; late transition, adopted EMR during last years of study; OR, odds ratio; CI, confidence interval; SE, standard error.

p < 0.05.

Additionally, after adjustment with covariates, a positive change in kappa was seen in sites with continuous use of EMRs with (β = 0.344, SE 0.13, p < 0.01) and without on-site mammography (β = 0.406, SE = 0.14, p < 0.01) compared with the reference site without EMRs or on-site mammography. In other words, the sites with EMRs had greater agreement among data sources than the sites without EMRs or on-site mammography (Table 5).

Table 5 shows other factors associated with agreement between self-report and medical record reports of mammography. Demographic factors, such as age at study entry, income, insurance type, and nativity, were not statistically significant predictors of the odds of agreement between medical record documentation and self-reported recent mammography use. Having a family history of breast cancer, however, decreased the odds of having agreement between self-report and medical record review documentation of mammography use (OR 0.43, 95% CI 0.23-0.82). There was only a 67% agreement between medical record review and self-report among women with a family history of breast cancer. The positive predictive value of self-report was 58% among women with a family history of breast cancer (n = 88) compared with 78% among women without a family history of breast cancer (n = 319); the negative predictive value of self-report was similar in the two groups (95% and 93%, respectively). The change in kappa statistic associated with family history of breast cancer was not statistically significant after adjustment for covariates. Although the odds of agreement between data sources associated with nativity were not statistically significant, the change in kappa statistic indicated a statistically significant decrease in data agreement associated with non-U.S. born status.

Discussion

Our study found that the agreement between the self-report and medical record estimates of recent mammography use was greater in sites with access to EMRs than those using paper records. The prevalence of mammography use estimated from medical records had the greatest agreement with women's self-report in sites with continuous availability of EMRs. The positive predictive value of self-report was upward of 86% in sites with continuous availability of EMRs; the positive predictive value of self-report was only 61% in the sites without EMRs. Moreover, the site with early transition to EMR during the study period had a higher positive predictive value of self-report than the site with late transition to EMR use. Taken together, the results suggest that sites with EMRs have a greater ability to capture the mammography screening experience reported by study participants than those without EMRs.

Consistent with prior literature, we found the prevalence of recent mammography use in our cohort, prior to intervention, varied greatly depending on whether self-report or medical record review estimated mammography use. When self-report was used to measure the prevalence of recent mammography use at baseline, most of our enrollment sites met Healthy People 2010 objectives by exceeding a 70% prevalence of use among women aged ≥40. When medical record reviews estimated recent mammography use at baseline, only one primary care site clearly met these objectives. Additionally, we found only modest evidence of telescoping as an explanation for these findings in our cohort, where 5% of participants reported having a mammogram more recently than was found in the medical record.

The discrepancies between self-report and medical record documentation of mammography use that have been described previously in the literature suggest positive predictive values for self-report ranging from approximately 60% to 95%.^8,15,16 African American women, in particular, are thought to overreport mammography use in self-report surveys.⁸ However, to our knowledge, our study is the first to discuss the greater discrepancy between self-report and medical record review in sites without EMRs. Our findings have implications for measuring the efficacy of mammography use interventions in underserved populations where the diffusion of information technology for record keeping is currently taking place.⁹ Although medical records are considered standard for assessing mammography use, primary care sites without EMR technology may report inordinately low rates of use because of documentation differences.¹⁷ Moreover, where EMR is adopted within sites, use trends may reflect measurement rather than intervention effects.

Our findings have important implications for clinical care. We found that sites without EMRs were less likely to have documentation of the mammograms that women believed they obtained. If women were correct and records are incomplete, it is possible that clinicians in sites without EMRs have less access to results and details for appropriate follow-up. Thus, if our findings are replicated, they raise the possibility that sites without electronic systems for documenting results have less complete data for measuring their success in promoting screening and providing clinical care to patients.

We note several limitations in our study. We were unable to capture other confounding information on site-level infrastructure that might affect the associations we report. For example, we were unable to gauge the effect of potentially varied methods for tracking mammography or integrating mammography reports into paper charts within sites (e.g., individual physicians' methods for managing results). Moreover, we were unable to gauge the presence or effect of billing/claims-based systems for tracking mammography use.¹⁸ We note the mixed experience using administrative claims data for tracking mammography use in underserved populations, with the greatest success in Medicare populations.^17,19,20 Indeed, Healthcare Effectiveness Data and Information Set (HEDIS) physician reporting for mammography uses a hybrid method that includes claims and medical record data in the reported measure (L. Anderson, National Center for Quality Assurance, personal communication, 2008). Additional research is needed to document the efficacy of claims-based systems for tracking mammography use among younger black women who are ineligible for Medicare or for uninsured women who have relied on free-care pools or out-of-pocket expenditures to pay for their mammograms.¹⁹

We were unable to evaluate the impact of specific features of EMRs that might explain greater agreement between self-report and medical record documentation where electronic systems are used. Improved documentation may facilitate the physician's review of the patient's screening history during the clinical visit, improve efficiency and effectiveness of communication about the need for mammography, and may enhance the patient's recall of mammography use. Therefore, in addition to providing documentation, because EMR systems also provide patient and provider reminders that reinforce the need for mammography use, women who attend sites with EMRs may have a greater ability to accurately report mammography use as a result of enhanced communication about regular screening. Future research should evaluate whether specific functions of EMRs (patient reminders, physician reminders, decision support) contribute to either the accuracy of patients' understanding and reporting of their mammography use or the prevalence of women's mammography use.

Importantly, the small sample size in some sites may limit the generalizability of our findings. Specifically, our findings on the differences in data agreement among sites that transitioned to EMR early during the study are based on small numbers. We find that trends in each site are concordant with findings in the literature, which show increased self-reporting of mammography compared with medical records, giving external validity to our results. However, replication in other settings would be required to ensure that these findings are generalizable to other primary care settings. Findings in one of the larger sites, the site with late transition to EMR, might reflect local implementation challenges. Indeed, although EMRs generally tend to improve documentation, challenges in implementation are known to cause unintended negative consequences. Consequences may include increasing time burden in entering information for documentation and, ironically, difficulties flagging patients due for screening where systems are not sufficiently equipped to provide decision support and reminder features to track health maintenance.^21,22 Although careful attention to implementation strategies is required to optimize the potential benefits of EMRs, we are unable to evaluate the effect of implementation challenges as a potential explanation for our results.

Additionally, we were unable to account fully for population differences that could affect the variation we report. For example, we were able to measure socioeconomic status (SES) by assessing annual household income; however, we did not have information on health literacy of patients that could affect self-reporting. Women with low health literacy may have overreported mammography use in our study because our survey did not use a lead-in stem to define mammography.²³ Previous research has noted that fewer black women with low SES report having a mammogram when a lead-in stem is used to give a description of a mammography procedure.²⁴ Our survey question without the lead-in stem was used systematically with all participants and across all sites with and without EMR availability. Moreover, no SES differences were noted across our study sites. Thus, it is likely that the discrepancy in data agreement among sites with and without EMRs is associated with features of EMRs rather than solely with the lack of a lead-in stem. It is possible that features of EMRs, including report generation and ability to print and mail reminders, may provide important educational messages. Additional research might explore the role of EMRs in enhancing health education and health literacy in underserved populations.

It is likely that urban clinics and community health centers have highly mobile populations with high patient turnover that present challenges for capturing their use of mammography that we were unable to evaluate fully. We have evaluated the role of nativity to attempt to understand if migration among black women affects the documentation of mammography use. Nativity did not consistently impact the relationships we report. Moreover, our longitudinal study previously showed that only a small percentage (3%) of our study participants moved from their enrollment site or from the city of enrollment during the study period (unpublished observations¹⁰). Nonetheless, we were unable to evaluate fully the challenges in capturing mammography use in urban clinics and community health centers with high mobile populations and high patient turnover. Finally, we restricted our inquiry to screening mammography and did not discuss systems for tracking abnormal results.

Despite these important limitations, this study has a number of strengths. We were able to evaluate the agreement between self-report and medical record reviews in primary care sites that had varying periods of time with access to EMRs and found a near dose-response relationship between the positive predictive value of self-report and access to EMRs for documenting mammography use. Second, the study data provided a unique opportunity to explore mammography documentation in a diverse population of black women at risk for underscreening. The data suggest an important infrastructure barrier to assessing the need for quality improvement in early breast cancer detection care, namely, a lack of access to technology that may more accurately track mammography use at women's primary care sites. Third, we were able to measure population characteristics that might affect women's ability to obtain and reliably report breast cancer detection practices, including the role of discrimination experiences and provider communication, that did not confound the associations we report. Finally, we are able to offer a hypothesis for previous findings that self-reported mammography use among black women may be overreported; documentation of black women's use patterns may be less systematically captured because of infrastructure differences in the primary care sites they attend. We found modest evidence for telescoping as an alternative hypothesis.⁸ We discovered the possibility of overreporting mammography use among women with a family history of breast cancer in our cohort.

In conclusion, we suggest that EMRs may provide more complete documentation of mammography use than paper records in a sample of primary care sites providing care to black women, as evidenced by a greater accuracy of mammography reporting among women in sites with EMRs than those without. In broad terms, our analysis underscores the importance of identifying technological and infrastructure characteristics that may affect the quality of care provided to underserved populations. In practical terms, instituting wide use of EMRs is likely to be a critical step in the infrastructure building needed to improve quality in mammography use for black women in underserved communities. At a minimum, intervention studies that seek to improve use of mammography in underserved communities should consider the mode of chart documentation used to track mammography use and seek supplemental data as appropriate.

Footnotes

Acknowledgments

We thank the members of the REACH Coalition, Dr. Barbara Ferrer, Dr. Snehal Shah, Dr. David Bates, Mr. Matt Fishman, and Ms. Wanda McClain, for their support and critique of this work. We thank the participating health centers and hospitals, REACH staff, and Ms. Amanda Mitchell at the Center for Community Health and Health Equity for her administrative assistance in the preparation of this article. The study was supported by a REACH 2010 grant from the U.S. Centers for Disease Control and Prevention. Additionally, the Center for Community Health and Health Equity and The 2006 Miles and Eleanor Shore Minority Faculty Development Award supported C.C. during the writing of this work. The financial sponsors of the research have not contributed to the design, interpretation, or manuscript preparation for this study.

Disclosure Statement

The authors have no conflicts of interest to report.

References

Ries

LAG

, Melbert

, Krapcho

et al. SEER Cancer Statistics Review, 1975–2004. Bethesda, MD: National Cancer Institute, 2007.

Bigby

, Ko

, Johnson

, David

, Ferrer

. A community approach to addressing excess breast and cervical cancer mortality among women of African descent in Boston. Public Health Rep, 2003; 118:338–347.

Use of mammograms among women aged greater or equal to 40 years: United States, 2000–2005. MMWR. 2007; 56:49–51.

Blackman

, Masi

. Racial and ethnic disparities in breast cancer mortality: Are we doing enough to address the root causes? J Clin Oncol, 2006; 24:2170–2178.

Bates

, Ebell

, Gotlieb

, Zapp

, Mullins

. A proposal for electronic medical records in U.S. primary care. J Am Med Inform Assoc, 2003; 10:1–10.

Soto

, Kleinman

, Simon

. Quality and correlates of medical record documentation in the ambulatory care setting. BMC Health Serv Res, 2002; 2:22–xx.

Baron

. Quality improvement with an electronic health record: Achievable, but not automatic. Ann Intern Med, 2007; 147:549–552.

Rauscher

, Johnson

, Cho

, Walk

. Accuracy of self-reported cancer-screening histories: A meta-analysis. Cancer Epidemiol Biomarkers Prev, 2008; 17:748–757.

Shields

, Shin

, Leu

et al. Adoption of health information technology in community health centers: Results of a national survey. Health Aff (Millwood), 2007; 26:1373–1383.

10.

Clark

, Baril

, Kunicki

, Johnson

, Soukup

, Ferguson

, Lipsitz

, Bigby

. Addressing social determinants of health to improve access to early breast cancer detection: Results of the Boston Reach 2010 Women's Health Demonstration Project. J Women's Health, 2009; 18:677–690.

11.

The Certification Commission for Healthcare Information Technology. Ambulatory certification criteria: 2008 final criteria. 2008.

12.

Coughlin

, Pickle

, Goodman

, Wilkens

. The logistic modeling of interobserver agreement. J Clin Epidemiol, 1992; 45:1237–1241.

13.

Lipsitz

, Williamson

, Klar

, Ibraham

, Parzen

. A simple method for estimating a regression model for kappa between a pair of raters. J R Statist Soc Series A, 2001; 164:449–465.

14.

McGinn

, Wyer

, Newman

, Keitz

, Leipzig

, For

. Tips for learners of evidence-based medicine: 3. Measures of observer variability (kappa statistic) Can Med Assoc J, 2004; 171:1369–1373.

15.

Caplan

, McQueen

, Qualters

, Leff

, Garrett

, Calonge

. Validity of women's self-reports of cancer screening test utilization in a managed care population. Cancer Epidemiol Biomarkers Prev, 2003; 12:1182–1187.

16.

McGovern

, Lurie

, Margolis

, Slater

. Accuracy of self-report of mammography and Pap smear in a low-income urban population. Am J Prev Med, 1998; 14:201–208.

17.

Armstrong

, Long

, Shea

. Measuring adherence to mammography screening recommendations among low-income women. Prev Med, 2004; 38:754–760.

18.

Holt

, Franks

, Meldrum

, Fiscella

. Mammography self-report and mammography claims: Racial, ethnic, and socioeconomic discrepancies among elderly women. Med Care, 2006; 44:513–518.

19.

Mouchawar

, Byers

, Warren

, Schluter

. The sensitivity of Medicare billing claims data for monitoring mammography use by elderly women. Med Care Res Rev, 2004; 61:116–127.

20.

Smith-Bindman

, Quale

, Chu

, Rosenberg

, Kerlikowske

. Can Medicare billing claims data be used to assess mammography utilization among women ages 65 and older? Med Care, 2006; 44:463–470.

21.

Ash

, Berg

, Coiera

. Some unintended consequences of information technology in health care: The nature of patient care information system-related errors. J Am Med Inform Assoc, 2004; 11:104–112.

22.

Crosson

, Stroebel

, Scott

, Stello

, Crabtree

. Implementing an electronic medical record in a family medicine practice: Communication, decision making, and conflict. Ann Fam Med, 2005; 3:307–311.

23.

Fiscella

, Franks

, Meldrum

. Estimating racial/ethnic disparity in mammography rates: It all depends on how you ask the question. Prev Med, 2004; 39:399–403.

24.

Siegel

, Qualters

, Mowery

, Campostrini

, Leutzinger

, McQueen

. Subgroup-specific effects of questionnaire wording on population-based estimates of mammography prevalence. Am J Public Health, 2001; 91:817–820.