Matern Child Health J DOI 10.1007/s10995-014-1461-8
Can an Electronic Health Record System be Used for Preconception Health Optimization? Heather Straub • Marci Adams • Richard K. Silver
Ó Springer Science+Business Media New York 2014
Abstract To explore the potential of an integrated outpatient electronic health record (EHR) for preconception health optimization. An automated case-finding EHR-derived algorithm was designed to identify women of child-bearing age having outpatient encounters in an 85-site, integrated health system. The algorithm simultaneously cross-referenced multiple discrete data fields to identify selected preconception factors (obesity, hypertension, diabetes, teratogen use including ACE inhibitors, multivitamin supplementation, anemia, renal insufficiency, untreated sexually transmitted infection, HIV positivity, and tobacco, alcohol or illegal drug use). Surveys were mailed to a random sample of patients to obtain their self-reported health profiles for these same factors. Concordance was assessed between the algorithm output, survey results, and manual data abstraction. Between 8/20102/2012, 107,339 female outpatient visits were identified, from which 29,691 unique women were presumed to have childbearing potential. 19,624 (66 %) and 8,652 (29 %) had 1 or C2 health factors, respectively while only 1,415 (5 %) had Presented in part as a poster at the 33rd Annual Meeting of the Society for Maternal-Fetal Medicine, San Francisco, CA, Feb 11–16, 2013.
Electronic supplementary material The online version of this article (doi:10.1007/s10995-014-1461-8) contains supplementary material, which is available to authorized users. H. Straub (&) M. Adams R. K. Silver Division of Maternal Fetal Medicine, NorthShore University HealthSystem, 2650 Ridge Ave, Walgreens Building, Suite 1507, Evanston, IL 60201, USA e-mail: [email protected] M. Adams e-mail: [email protected] R. K. Silver e-mail: [email protected]
none. Using the patient survey results as a reference point, health-factor agreement was similar comparing the algorithm (85.8 %) and the chart abstraction (87.2 %) results. Incorrect or missing data entries in the EHR encounters were largely responsible for discordances observed. Preconception screening using an automated algorithm in a system-wide EHR identified a large group of women with potentially modifiable preconception health conditions. The issue most responsible for limiting algorithm performance was incomplete point of care documentation. Accurate data capture during patient encounters should be a focus for quality improvement, so that novel applications of system-wide data mining can be reliably implemented. Keywords Preconception health Electronic health record Clinical data mining Health information accuracy
Introduction Up to 30 % of all pregnancies in the United States develop complications that could be mitigated by improvements in preconception health [1]. Examples include optimizing treatment for selected medical conditions such as diabetes, for which the perinatal consequences of poor glucose control include higher rates of miscarriage and fetal malformations [2]. Opportunities exist for pregnancy improvement related to ‘‘sins of omission.’’ Preconception vitamin supplementation is a relevant example, as national patient education campaigns, [3] food fortification efforts [4] and other public health strategies [5] have had only modest impact on adherence to vitamin supplementation and folate status among reproductive-age women [6, 7]. Part of the problem that limits timely preconception intervention is the high prevalence of unplanned pregnancy,
123
Matern Child Health J
estimated at approximately 50 % in the U.S. [8]. However, since many American women interface with the health system on a regular basis for other reasons, there could be opportunities for preconception health assessment even if a patient is not planning to conceive in a given timeframe. According to the Agency for Healthcare Research and Quality, of 55,507 patients who took their medical expenditure survey, 87.9 % of women between the ages of 18–44, had seen a healthcare provider at least once in the prior year [9]. However, these visits were not necessarily for the purpose of pregnancy preparation and it is unclear if preconception topics were discussed. Even when patient encounters involve an obstetric provider as part of wellwoman care, preconception topics are not routinely covered [10]. A more effective strategy is needed in which preconception health is both assessed and modified during an unrelated patient care encounter. To be effective, such an initiative should meet patients where they are in the care continuum rather than add additional encounters to their schedules or to the system. Further, we hypothesize that the electronic health record (EHR) could enable such a strategy by embedding preconception health assessment into the routine conduct of outpatient care rather than by trying to orchestrate a specific ‘‘just in time’’ pre-pregnancy visit for many thousands of women. Taking the EHR beyond being an expensive electronic copy of a paper chart is still an area requiring investigation and innovation. Automation of screening practices using an EHR has been shown to improve compliance with recommended guidelines [11, 12] and increase the efficiency of chart review for focused research [13]. Likewise, electronic trigger tools have been developed to identify patients with undiagnosed or undocumented medical conditions such as hypertension [14] or to analyze sentinel lab values to monitor potential adverse drug events [15]. The EHR has also been used to accomplish automated risk stratification for cardiac disease and encourage corresponding behavioral interventions among those patents so identified [16]. To test our hypothesis that an enterprise-wide EHR can be used as a first step toward preconception health optimization, we developed a novel case-finding algorithm designed to flag women of child-bearing potential with modifiable health factors. In this report we evaluate the feasibility and reliability of automated preconception health status assessment in comparison to direct patient survey and physician chart abstraction.
Materials and Methods A multi-dimension algorithm spanning several enterprise data warehouse (EDW) data marts was designed to assess
123
women for the potential to become pregnant and report on preconception health factors that were derived from a review of relevant health guidelines for pre-pregnancy care. Preconception health factors were selected for inclusion in this study only if a direct link with pregnancy complications was established in the literature AND adverse pregnancy outcomes could be mitigated by prepregnancy interventions [1, 17]. Those factors included also had to be amenable to EHR detection (i.e., recorded in discrete data fields). The following factors were selected: morbid obesity, hypertension, diabetes, teratogen use, multivitamin supplementation, anemia, renal insufficiency, untreated sexually transmitted diseases, HIV, alcohol, tobacco and substance use (Table 1). Other factors thought to be relevant to preconception health were not included because of their low frequency in our population (e.g. maternal phenylketonuria) or non-documentation in discrete data fields (e.g., environmental exposures). Obstetric risk factors (e.g., prior preterm birth, prior stillbirth, preeclampsia and others) were not included because with few exceptions, these conditions would not warrant a change in medical care prior to conception. System-wide EHR clinical encounter data were uploaded daily to the EDW (data flows from EPIC to Clarity to EDW Oracle database to the IBM Cognos Reporting Tool), to which the case-finding algorithm was applied with the following inclusion criteria: (1) female patients, ages 12–45; (2) no contemporaneous pregnancy episode in the system; (3) no active birth control prescriptions or longterm reversible contraceptive use in their medication or problem lists; and, (4) no surgical procedure or historical codes indicative of prior sterilization or hysterectomy. Women who were actively using contraception were excluded because they were considered not to be of childbearing potential at time of analysis. Since behavioral and non-prescription birth control methods were not assessed in any discrete EHR fields, the algorithm did not account for use of these lower efficacy contraceptives [18]. We acknowledged that contraceptive status can change over time and recognized that even women using contraception can conceive without planning to do so. The algorithm next checked the status for each preconception health factor considered to be either behavioral/ nutritional (e.g., obesity, multivitamin supplementation, alcohol use, tobacco use, or illegal drug use) or related to a medical disease/medication (e.g. hypertension, uncontrolled diabetes, teratogen exposure including ACE inhibitor use, anemia, renal insufficiency, untreated sexually transmitted diseases or HIV positivity). The computer program simultaneously cross-referenced all problem list entries, all medical history fields, all diagnoses, selected clinical measures (e.g., height, weight and blood pressure), selected laboratory assays (e.g., hemoglobin, serum
Matern Child Health J
glucose, serum creatinine, urine toxicology), all medication orders, and selected ICD-9 codes. The algorithm logic was set so that a variable positive by any aspect of the query categorized the patient as positive for that health factor in
Table 1 Algorithm definitions for preconception health factors using the electronic health record (EHR) data eSlements Preconception health factor
Definition applied by the case-finding algorithm and data sources queried within the EHR
Alcohol use
Documentation provided in social history in discrete fields
Anemia
Hemoglobin \ 10 g/dL or Hematocrit \ 30 % in laboratory results
ACE inhibitor use
Documentation of ACE inhibitor in active medication list
Hypertension
Documentation as either a diagnosis, an element of medical history or by a set of ICD 9 codes
Illegal drug use
Documentation in medical or social history, a set of ICD 9 codes or by laboratory results such as positive urine toxicology for cocaine, marijuana or opiates
No Vitamin supplementation
No multivitamins identified in active medication list
Obesity
Weight (kg)/height (m)2 from vital signs section of encounter
Positive HIV
Documentation as either a diagnosis, a set of ICD 9 codes or by positive laboratory serology
Renal insufficiency
N: 107, 339
Documentation as either a diagnosis, a set of ICD 9 codes or in laboratory results by serum creatinine [ 1.5 mg/dL
Teratogen exposure
Documentation of pre-specified teratogenic medications in active medication lista
Tobacco use
Documentation provided in social history or a set of ICD 9 codes
Sub-optimally controlled diabetes
Documentation in medical history, a set of ICD 9 codes or by Hgb A1c [ 6 % or laboratory result of serum glucose [200 mg/dL
Untreated Chlamydia
Positive laboratory results or set of ICD 9 codes without documentation of a corresponding macrolide or tetracycline in the medication list
Untreated gonorrhea
Positive laboratory results or a set of ICD 9 codes without documentation of a corresponding cephalosporin in medication list
Untreated syphilis
order to err on the side of over-identification rather than under-ascertainment. This study was performed according to prevailing ethical principles. We obtained institutional review board approval, and waivers of written consent and authorization were obtained to contact a subset of identified patients to participate in a survey. The survey instrument was designed to learn the status of the preconception health factors from the patients’ perspective without their knowledge of the algorithm results. Survey questions were initially reviewed for provider and patient face validity by obstetricians and a convenience sample of non-pregnant women. A random number was assigned to each of the potential survey subjects and patients numbered 1–1,000 were solicited to participate. If a patient did not respond within 1 month a reminder postcard was sent, and nonresponders at 2 months received a second mailed survey. Completed surveys triggered a detailed review of the corresponding patient EHR masked to the survey responses and the algorithm results. Comparisons were made between the algorithm, survey results and chart reviews for all health factors. Concordance was defined as the percentage of women who had agreement between their self-reported health factors by survey and the corresponding algorithm and manual chart
n: 29, 691
n: 1,000
n: 664
n: 336
n: 109
n: 83
Positive laboratory results or a set of ICD 9 codes without documentation of a corresponding penicillin or acceptable alternative in medication list
a
Teratogen medications categories included anticonvulsants, benzodiazepines, Coumadin/warfarin, 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors (statins), lithium, tetracyclines, and triptan medications
n: 141
Fig. 1 Flow diagram of surveys sent and analyzed
123
123
0 (0)
Sub-optimally controlled diabetes
0 (0)
2 (1.4) 1 (0.7)
4 (2.8)
12(8.5)
13 (9.2)
15 (10.6)
12 (8.5)
21 (14.9)
NA
83 (58.9)
106 (75.2)
0:141
0:139
0:140
7:133
12:110
1:136
3:130
22:113
12:118
32:79
141 (100)
139 (98.6)
140 (99.3)
140 (99.3)
122 (86.5)
137 (97.2)
133 (94.3)
135 (95.7)
130 (92.2)
111 (78.7)
Total n (%)
NA
0 (0)
0 (0)
0.93
0.48
-0.35
0.40
0.85
0.64
0.54
Kappac
0:140
0:137
1:129
7:127
8:110
1:128
3:116
NA
16:57
32:34
Pos:Neg n:n
140 (99.3)
137 (97.2)
130 (92.2)
134 (95.0)
118 (83.7)
129 (91.5)
119 (84.4)
NA
73 (51.8)
66 (46.8)
Total n (%)
0.50
0.49
0.52
0.71
0.11
0.48
0.42
NA
0.37
0.30
Kappa
Between algorithm and survey
0:140
2:137
0:129
7:128
5:112
3:128
7:120
NA
16:56
56:30
Pos:Neg n:n
140 (99.3)
139 (98.6)
129 (91.5)
135 (95.7)
117 (83.0)
131 (92.9)
127 (90.1)
NA
72 (51.1)
86 (61.0)
Total n (%)
0 (0)
0.66
0 (0)
0.68
0.20
0.27
0.46
NA
0.14
0.27
Kappa
Between survey and chart review
Pos positive, Neg negative, NA not asked
Kappa coefficients were calculated where j = 0 reflects an agreement purely due to chance and a j B 0.40 reflects a slight to fair agreement, and a j [ 0.40 reflecting a moderate to substantial agreement
c
b
Renal insufficiency, positive HIV status and untreated STD were not identified from the survey nor found in the chart abstraction or algorithm results Concordance defined as combined positive–positive and negative–negative health factor comparisons; listed in the first column as positive–positive:negative–negative and in the second column as total concordance (combined positive–positive and negative–negative)
a
0 (0)
ACE inhibitor use
0 (0)
1(0.7)
Anemia
2 (1.4)
7 (5.0)
1 (0.7)
Illegal drug use
7 (5.0)
8 (5.7)
7 (5.0)
Tobacco use
24 (17.0)
Hypertension
26 (18.4)
Obesity
18 (12.8)
19 (13.5)
17 (12.1)
Multivitamin use
60 (42.6)
Medical conditions/medications Teratogen exposure 24 (17.0)
33 (23.4)
Pos:Neg n:n
Between algorithm and chart review
Survey health factor
Algorithm health factor
Chart abstraction health factor
Concordanceb
Frequency n (%)
Alcohol use
Behavioral/nutritional
Health factor
Table 2 Frequency and concordance of preconception health factors derived from the algorithm, chart abstraction and patient surveya (n = 141)
Matern Child Health J
Matern Child Health J
review results. Chi squared analyses and Fisher’s exact tests were used to compare demographic and health factors between data gathering methods. Kappa coefficients were calculated where j = 0 reflects an agreement purely due to chance and a j B 0.40 reflects a slight to fair agreement, and a j [ 0.40 reflects a moderate to substantial agreement [19]. The Kappa coefficient is limited by prevalence, so positive–positive, negative–negative and total correlations were also reported. A p value of \0.05 was considered statistically significant.
Results Between 8/1/2010 and 2/29/2012 automated analysis of 107,339 outpatient encounters yielded 29,691 women who were of child-bearing potential. Patient ages were evenly distributed with 30 % between ages 18–24; 34 % between ages 25–35; and 36 % between ages 36–45. Over half (52.6 %) of these patients had C10 outpatient visits during the study period, suggesting that the health system and its providers were likely the primary source for their care. A majority of women had at least one preconception health factor (n = 19,624 or 66 %) while 8,652 (29 %) had C2 health factors and only 1,415 (5 %) had no health factors as assessed by the algorithm. Of the 1,000 patients randomly selected to receive a survey, 336 (33.6 %) were returned by the postal service (Fig. 1). After excluding wrong addresses with no forwarding address (n = 109), wrong addresses re-sent to forwarding address (n = 83) and 3 requested exclusions, 141 surveys were available for analysis. Preconception health factors identified via the algorithm, the chart abstraction and the patient surveys are noted in Table 2. Patients who returned the survey were similar in most respects to non-participants except that they were slightly older on average (32.7 ± 8.4 vs. 30.6 ± 8.3, p = 0.006) and more likely to have had increased contact with the health system during the study timeframe (C10 outpatient visits; 65.2 vs. 50.5 %, p = 0.001). The health factor profile derived from the algorithm among patients who returned the survey compared to those patients who did not was only different for one variable—alcohol consumption (23.4 vs. 12.0 %, respectively, p \ 0.001). The algorithm misidentified 43 women(30.5 %) as being of child-bearing potential by study definition when in fact, they were either using contraception (n = 39); already pregnant but receiving care outside of health system (n = 3); or not seeking pregnancy with their same-gendered partner (n = 1). Compared to the patient survey results, the algorithm and the chart review both underreported a majority of the preconception health factors secondary to incomplete EHR documentation (Table 2). The
algorithm results demonstrated high overall concordance (95.1 %) when compared to chart abstraction, suggesting that the case-finding logic accurately captured and interpreted the available discrete EHR entries made by providers. The next highest concordance was observed between patient survey results and chart abstraction (87.2 %), with similar concordance between survey results and the algorithm (85.8 %). Hypertension had the highest overall concordance between the algorithm and chart review (j = 0.93), the algorithm and survey (j = 0.71), and between the chart review and survey (j = 0.68). Where the algorithm performed less well compared to the survey (e.g., alcohol use (j = 0.27), the EHR chart appeared incomplete or incorrect.
Conclusions Our EHR algorithm identified almost 30,000 women seen as outpatients over 18 months, the majority of whom had potentially modifiable preconception health factors. When compared to patient reporting via our survey, the EHR chart and the algorithm underestimated health profiles but not by large margins. Missing or incomplete data entry in the electronic patient encounters was the primary explanation for this finding. Nonetheless, the algorithm performed well as indicated by its ‘‘internal validity’’ with 95.1 % average concordance compared to chart abstraction. The accuracy of the algorithm suggests that it could be useful in identifying at-risk women and flagging them for preconception counseling within the constraints of complete and accurate EHR documentation by providers. Both the algorithm and chart review appeared to underreport health factors when compared to the patient survey. Whether providers did not ask these questions, whether the answers were not recorded or if responses to selected health questions were misrepresented by patients is unclear. When Killeen and colleagues looked at concordance between self reported tobacco, alcohol and drug use compared to health record extraction they found discordance related to type of medical visit and insurance status [20]. Since all encounters queried were outpatient visits and insurance status was variable in our cohort, selective reporting by patients may have been at play. However, the under representation of the prevalence of alcohol and tobacco use by the algorithm and the chart review raises further concerns associated with incomplete data entry. Incomplete data entry can also have a potentially negative impact on patient safety well beyond our theoretic application in preconception health [21]. Missing data is also a common pitfall in EHR research [22] which we attempted to overcome by using multiple fields per factor to capture information entered in one of several possible
123
Matern Child Health J
locations. While we did not data mine free text entries, the efficiency of our algorithm suggests that adding this technology would provide only incremental improvement at best, while the potential downside would be significantly increased computer runtime (for the processing of 18 months of outpatient encounters our algorithm required approximately 24 h). Skeptics might fault the algorithm for misidentifying *30 % of the patients being of child-bearing potential owing to missing EHR data. This missing data rate is comparable to other studies which have reported ranges from *20–50 % [23–25] and we would assert that the algorithm is intended as a safety-net from which confirmation of health status would still be required as the next logical step before any patient-specific interventions would be undertaken. It appears that 29,691 women within our healthcare system were of child bearing potential with 95 % having at least one of the identified preconception health factors. Even if 30 % of these women were erroneously identified by our algorithm as being at-risk for pregnancy, this would still leave over 20,000 women who could potentially benefit from an intervention. We have no way of determining if our algorithm identified all women of child-bearing potential during this period (i.e., falsenegatives); however, we were conservative in the design of the algorithm to lean towards over ascertainment which represents a significant improvement above our baseline of usual care. The costs of over-identification would depend on the designed intervention which could range from an automated alert system in the EHR (inexpensive to deploy and maintain) to direct outreach by clinical personnel (requiring significant resources). Rakotz and colleagues designed an algorithm to screen for undiagnosed hypertension which identified 1,600 patients of which, 475 came in for confirmatory blood pressure measurements [14]. While they found that less than half (38 %) of patients flagged as hypertensive were confirmed to be truly hypertensive, the cost of their confirmatory clinical evaluations among normotensive cases must be considered in the context of potential health savings for hypertensive subjects who had not been identified previously and were atrisk for additional morbidity and cost without treatment. Patient recall bias could have impacted the accuracy of the survey results which in turn, could also have influenced concordance calculations with the algorithm and chart abstraction datasets. To try and limit this confounder, the interval between the actual patient encounters and the completion of the survey was within 12-months in a majority of subjects and the types of health questions posed were largely categorical. We also acknowledge that our case-finding algorithm remains vulnerable to errors in primary data entry. However, the 87.2 % average concordance rate comparing the survey to the chart abstraction
123
was at least comparable to best performance reported in another study of electronic data mining [23] and superior when compared to an observed accuracy rate of 30.7 % in a study reviewing automated data retrieval of 5,135 patient records for surgically related clinical questions. [26] In the future, EHR data accuracy may improve in response to meaningful use legislation linking complete discrete data field documentation to reimbursement [27]. Because comparative effectiveness research in preconception care is not yet available [28, 29] the proper strategy to modify risk for the women identified through our algorithm is unknown. Potential interventions could include communicating with the patients’ primary care providers via electronic physician alerts or computer-based decision tools, which have been shown to improve compliance with clinical protocols [30] and decrease teratogenic medication exposures in women without documented contraception [31]. When these electronic solutions are well designed, such innovations can have high acceptance rates from providers [32]. Alternatively, direct communication via a secure internet portal could invite the patient to be assessed for an undiagnosed or incompletely treated health condition or simply provide education on the importance of health management (e.g., vitamin supplementation). A pilot study looking at cardiovascular risk stratification has found that automated, personally targeted intervention strategies can work [16] and we have undertaken a similar study to improve preconception diabetes management. At present, over 200,000 patients use our EHR internet portal for results review, appointment scheduling and direct communication with their physicians. Since almost all (95.7 %) of the patients we surveyed reported being interested in receiving information about improving health prior to pregnancy, it would be reasonable to test this confidential web portal for such communications since it is already connected to the EHR system and could be an attractive platform for intervention. Further research is needed to determine whether preconception health improvements as reported by others [33] can actually be facilitated by first relying on an automated EHR screening algorithm similar to the one we have studied. Acknowledgments Thank you to Tom East, PhD, Doris Ng and Lourdes Link for their guidance on study design, algorithm development and data processing. This project was funded through intramural support. Conflict of interest None of the authors have a conflict of interest.
References 1. Centers for Disease Control and Prevention. (2006). Recommendations for improving preconception health and health care: United States: A report of the CC/ATSDR Preconception Care
Matern Child Health J
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Workgroup and the Select Panel on Preconception Care. MMWR Morbidity and Mortality Weekly Report, 55, 1–23. Kitzmiller, I. L., Buchanan, T. A., Kjos, S., Combs, C. A., & Ratner, R. (1996). Preconception care of diabetes, congenital malformations, and spontaneous abortions (technical review). Diabetes Care, 19, 514–541. Centers for Disease Control and Prevention. (1992). Recommendations for the use of folic acid to reduce the number of cases of spina bifida and other neural tube defects. MMWR Morbidity and Mortality Weekly Report, 41, 1–6. Honein, M. A., Paulozzi, L. J., Matthews, T. J., Erickson, J. D., & Wong, L. C. (2001). Impact of folic acid fortification of the US food supply on the occurrence of neural tube defects. JAMA, 285(23), 2981–2986. Rose, D. (2010) A history of the March of Dimes. March of Dimes Archives (Accessed 13 March, 2013. http://www.march ofdimes.com/mission/history_indepth.html). Olney, R. S., & Mulinare, J. (2002). Trends in neural tube defect prevalence, folic acid fortification, and vitamin supplement use. Seminars in Perinatology, 26(4), 277–285. Centers for Disease Control and Prevention. (1999). Knowledge and use of folic acid by women of childbearing age-United States, 1995 and 1998. MMWR Morbidity and Mortality Weekly Report, 48, 325–327. Finer, L. B., & Zolna, M. R. (2011). Unintended pregnancy in the United States: Incidence and disparities, 2006. Contraception, 84(5), 478–485. Taylor, A. K., Larson, S., & Correa-de-Araujo, R. (2006). Women’s healthcare utilization and expenditures. Women’s Health Issues, 16, 66–79. Morgan, M. A., Anderson, B. L., Lawrence, H., & Schulkin, J. (2012). Well-woman care among obstetrician-gynecologists: Opportunity for preconception care. Journal of Maternal-Fetal and Neonatal Medicine, 25(6), 595–599. Sankilampi, U., Saari, A., Laine, T., Miettinen, P. J., & Dunkel, L. (2013). Use of electronic health records for automated screening of growth disorders in primary care. JAMA, 310(10), 1071–1072. Anand, V., Carroll, A. E., & Downs, S. M. (2012). Automated primary care screening in pediatric waiting rooms. Pediatrics, 129(5), e1275–e1281. Seyfried, L., Hanauer, D. A., Nease, D., Albeiruti, R., Kavanagh, J., & Kales, H. C. (2009). Enhanced identification of eligibility for depression research using an electronic medical record search engine. International Journal Medical Informatics, 78(12), e13–e18. Rakotz, M. K., Ross, R. E., Robicsek, A., Konchak, C., Gavagan, T., & Ewigman, B. (2012). Searching for undiagnosed hypertension: A case study in the translation of quality research into operations. In International Health Care Quality Meeting, Paris, France, April 2012. Brenner, S., Detz, A., Lopez, A., Horton, C., & Sarkar, U. (2012). Signal and noise: applying a laboratory trigger tool to identify adverse drug events among primary care patients. BMJ Quality & Safety, 21(8), 670–675. Jones, J. B., Shah, N. R., Bruce, C. A., & Stewart, W. F. (2011). Meaningful use in practice using patient-specific risk in an electronic health record for shared decision making. American Journal of Preventive Medicine, 40(5 Suppl 2), S179–S186. Jack, B. W. Atrash, H., Coonrod, D. V., Moos, M. K., O’Donnell, J., & Johnson, K. (2008). The clinical content of preconception care: An overview and preparation of this supplement. American Journal of Obstetrics and Gynecology, 199(6 Suppl 2), S266–S279. Mansour, D., Inki, P., & Gemzell-Danielsson, K. (2010). Efficacy of contraceptive methods: A review of the literature. The
19. 20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
European Journal of Contraception and Reproductive Health Care, 15(1), 4–16. Viera, A. J., & Garrett, J. M. (2005). Understanding interobserver agreement: The kappa statistic. Family Medicine, 37(5), 360–363. Killeen, T. K., Brady, K. T., Gold, P. B., Tyson, C., & Simpson, K. N. (2004). Comparison of self-report versus agency records of service utilization in a community sample of individuals with alcohol use disorders. Drug and Alcohol Dependence, 73, 141–147. Institute of Medicine. (2001). Building organizational supports for change. In Institute of Medicine (Ed.), Crossing the quality chasm: A new health system for the 21st century (pp. 111–144). Washington, DC: National Academy Press. Terry, A. L., Chevendra, V., Thind, A., Stewart, M., Marshall, J. N., & Cejic, S. (2010). Using your electronic medical record for research: A primer for avoiding pitfalls. Family Practice, 27, 121–126. Prather, J. C., Lobach, D. F., Goodwin, L. K., Hales, J. W., Hage, M. L., & Hammond, W. E. (1997). Medical data mining: Knowledge discovery in a clinical data warehouse. In Proceedings of the AMIA annual fall symposium, p. 101–105. Botsis, T., Hartvigsen, G., Chen, F., Weng, C. (2010). Secondary use of EHR: data quality issue and informatics opportunities. In AMIA summits on translational science proceedings, p. 1–5. Newgard, C. D., Zive, D., Jui, J., Weathers, C., & Daya, M. (2012). Electronic versus manual data processing: Evaluating the use of electronic health records in out-of-hospital clinical research. Academic Emergency Medicine, 19(2), 217–227. Stein, H. D., Nadkarni, P., Erdos, J., & Miller, P. L. (2000). Exploring the degree of concordance of coded and textual data in answering clinical queries from a clinical data repository. Journal of the American Medical Informatics Association, 7(1), 42–54. Hasan, S., & Padman, R. (2006) Analyzing the effect of data quality on the accuracy of clinical decision support systems: A computer simulation approach. In AMIA annual symposium proceedings, p. 324–328. Zywiak, W., Metzer, J., & Mann, M. (2010). Meaningful use for eligible professionals: The top ten challenges. (Accessed March 13, 2013. http://www.csc.com/health_services/insights/44165-mean ingful_use_for_eligible_professionals_the_top_ten_challenges). Salihu, H. M., Salinas, A., & Mogos, M. (2013). The missing link in preconceptional care: The role of comparative effectiveness research. Maternal and Child Health Journal, 17(5), 776–782. Whitworth, M., & Dowswell, T. (2009). Routine pre-pregnancy health promotion for improving pregnancy outcomes. Cochrane Database of Systematic Reviews, 2009(4). doi:10.1002/14651858. CD007536.pub2. Garg, A. X., Adhikari, N. K., McDonald, H., Rosas-Arellano, M. P., Devereaux, P. J., Beyene, J., et al. (2005). Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: A systematic review. JAMA, 293(10), 1223–1238. Schwarz, E. B., Parisi, S. M., Handler, S. M., Koren, G., Cohen, E. D., Shevchik, G. J., et al. (2012). Clinical decision support to promote safe prescribing to women of reproductive age: A cluster-randomized trial. Journal of General Internal Medicine, 27(7), 831–838. Shannon, G. D., Alberg, C., Nacul, L., & Pashayan, N. (2013). Preconception healthcare and congenital disorders: systematic review of the effectiveness of preconception care programs in the prevention of congenital disorders. Maternal and Child Health Journal. doi:10.1007/s10995-013-1370-2.
123
Can an Electronic Health Record System be Used for Preconception Health Optimization? Heather Straub • Marci Adams • Richard K. Silver
Ó Springer Science+Business Media New York 2014
Abstract To explore the potential of an integrated outpatient electronic health record (EHR) for preconception health optimization. An automated case-finding EHR-derived algorithm was designed to identify women of child-bearing age having outpatient encounters in an 85-site, integrated health system. The algorithm simultaneously cross-referenced multiple discrete data fields to identify selected preconception factors (obesity, hypertension, diabetes, teratogen use including ACE inhibitors, multivitamin supplementation, anemia, renal insufficiency, untreated sexually transmitted infection, HIV positivity, and tobacco, alcohol or illegal drug use). Surveys were mailed to a random sample of patients to obtain their self-reported health profiles for these same factors. Concordance was assessed between the algorithm output, survey results, and manual data abstraction. Between 8/20102/2012, 107,339 female outpatient visits were identified, from which 29,691 unique women were presumed to have childbearing potential. 19,624 (66 %) and 8,652 (29 %) had 1 or C2 health factors, respectively while only 1,415 (5 %) had Presented in part as a poster at the 33rd Annual Meeting of the Society for Maternal-Fetal Medicine, San Francisco, CA, Feb 11–16, 2013.
Electronic supplementary material The online version of this article (doi:10.1007/s10995-014-1461-8) contains supplementary material, which is available to authorized users. H. Straub (&) M. Adams R. K. Silver Division of Maternal Fetal Medicine, NorthShore University HealthSystem, 2650 Ridge Ave, Walgreens Building, Suite 1507, Evanston, IL 60201, USA e-mail: [email protected] M. Adams e-mail: [email protected] R. K. Silver e-mail: [email protected]
none. Using the patient survey results as a reference point, health-factor agreement was similar comparing the algorithm (85.8 %) and the chart abstraction (87.2 %) results. Incorrect or missing data entries in the EHR encounters were largely responsible for discordances observed. Preconception screening using an automated algorithm in a system-wide EHR identified a large group of women with potentially modifiable preconception health conditions. The issue most responsible for limiting algorithm performance was incomplete point of care documentation. Accurate data capture during patient encounters should be a focus for quality improvement, so that novel applications of system-wide data mining can be reliably implemented. Keywords Preconception health Electronic health record Clinical data mining Health information accuracy
Introduction Up to 30 % of all pregnancies in the United States develop complications that could be mitigated by improvements in preconception health [1]. Examples include optimizing treatment for selected medical conditions such as diabetes, for which the perinatal consequences of poor glucose control include higher rates of miscarriage and fetal malformations [2]. Opportunities exist for pregnancy improvement related to ‘‘sins of omission.’’ Preconception vitamin supplementation is a relevant example, as national patient education campaigns, [3] food fortification efforts [4] and other public health strategies [5] have had only modest impact on adherence to vitamin supplementation and folate status among reproductive-age women [6, 7]. Part of the problem that limits timely preconception intervention is the high prevalence of unplanned pregnancy,
123
Matern Child Health J
estimated at approximately 50 % in the U.S. [8]. However, since many American women interface with the health system on a regular basis for other reasons, there could be opportunities for preconception health assessment even if a patient is not planning to conceive in a given timeframe. According to the Agency for Healthcare Research and Quality, of 55,507 patients who took their medical expenditure survey, 87.9 % of women between the ages of 18–44, had seen a healthcare provider at least once in the prior year [9]. However, these visits were not necessarily for the purpose of pregnancy preparation and it is unclear if preconception topics were discussed. Even when patient encounters involve an obstetric provider as part of wellwoman care, preconception topics are not routinely covered [10]. A more effective strategy is needed in which preconception health is both assessed and modified during an unrelated patient care encounter. To be effective, such an initiative should meet patients where they are in the care continuum rather than add additional encounters to their schedules or to the system. Further, we hypothesize that the electronic health record (EHR) could enable such a strategy by embedding preconception health assessment into the routine conduct of outpatient care rather than by trying to orchestrate a specific ‘‘just in time’’ pre-pregnancy visit for many thousands of women. Taking the EHR beyond being an expensive electronic copy of a paper chart is still an area requiring investigation and innovation. Automation of screening practices using an EHR has been shown to improve compliance with recommended guidelines [11, 12] and increase the efficiency of chart review for focused research [13]. Likewise, electronic trigger tools have been developed to identify patients with undiagnosed or undocumented medical conditions such as hypertension [14] or to analyze sentinel lab values to monitor potential adverse drug events [15]. The EHR has also been used to accomplish automated risk stratification for cardiac disease and encourage corresponding behavioral interventions among those patents so identified [16]. To test our hypothesis that an enterprise-wide EHR can be used as a first step toward preconception health optimization, we developed a novel case-finding algorithm designed to flag women of child-bearing potential with modifiable health factors. In this report we evaluate the feasibility and reliability of automated preconception health status assessment in comparison to direct patient survey and physician chart abstraction.
Materials and Methods A multi-dimension algorithm spanning several enterprise data warehouse (EDW) data marts was designed to assess
123
women for the potential to become pregnant and report on preconception health factors that were derived from a review of relevant health guidelines for pre-pregnancy care. Preconception health factors were selected for inclusion in this study only if a direct link with pregnancy complications was established in the literature AND adverse pregnancy outcomes could be mitigated by prepregnancy interventions [1, 17]. Those factors included also had to be amenable to EHR detection (i.e., recorded in discrete data fields). The following factors were selected: morbid obesity, hypertension, diabetes, teratogen use, multivitamin supplementation, anemia, renal insufficiency, untreated sexually transmitted diseases, HIV, alcohol, tobacco and substance use (Table 1). Other factors thought to be relevant to preconception health were not included because of their low frequency in our population (e.g. maternal phenylketonuria) or non-documentation in discrete data fields (e.g., environmental exposures). Obstetric risk factors (e.g., prior preterm birth, prior stillbirth, preeclampsia and others) were not included because with few exceptions, these conditions would not warrant a change in medical care prior to conception. System-wide EHR clinical encounter data were uploaded daily to the EDW (data flows from EPIC to Clarity to EDW Oracle database to the IBM Cognos Reporting Tool), to which the case-finding algorithm was applied with the following inclusion criteria: (1) female patients, ages 12–45; (2) no contemporaneous pregnancy episode in the system; (3) no active birth control prescriptions or longterm reversible contraceptive use in their medication or problem lists; and, (4) no surgical procedure or historical codes indicative of prior sterilization or hysterectomy. Women who were actively using contraception were excluded because they were considered not to be of childbearing potential at time of analysis. Since behavioral and non-prescription birth control methods were not assessed in any discrete EHR fields, the algorithm did not account for use of these lower efficacy contraceptives [18]. We acknowledged that contraceptive status can change over time and recognized that even women using contraception can conceive without planning to do so. The algorithm next checked the status for each preconception health factor considered to be either behavioral/ nutritional (e.g., obesity, multivitamin supplementation, alcohol use, tobacco use, or illegal drug use) or related to a medical disease/medication (e.g. hypertension, uncontrolled diabetes, teratogen exposure including ACE inhibitor use, anemia, renal insufficiency, untreated sexually transmitted diseases or HIV positivity). The computer program simultaneously cross-referenced all problem list entries, all medical history fields, all diagnoses, selected clinical measures (e.g., height, weight and blood pressure), selected laboratory assays (e.g., hemoglobin, serum
Matern Child Health J
glucose, serum creatinine, urine toxicology), all medication orders, and selected ICD-9 codes. The algorithm logic was set so that a variable positive by any aspect of the query categorized the patient as positive for that health factor in
Table 1 Algorithm definitions for preconception health factors using the electronic health record (EHR) data eSlements Preconception health factor
Definition applied by the case-finding algorithm and data sources queried within the EHR
Alcohol use
Documentation provided in social history in discrete fields
Anemia
Hemoglobin \ 10 g/dL or Hematocrit \ 30 % in laboratory results
ACE inhibitor use
Documentation of ACE inhibitor in active medication list
Hypertension
Documentation as either a diagnosis, an element of medical history or by a set of ICD 9 codes
Illegal drug use
Documentation in medical or social history, a set of ICD 9 codes or by laboratory results such as positive urine toxicology for cocaine, marijuana or opiates
No Vitamin supplementation
No multivitamins identified in active medication list
Obesity
Weight (kg)/height (m)2 from vital signs section of encounter
Positive HIV
Documentation as either a diagnosis, a set of ICD 9 codes or by positive laboratory serology
Renal insufficiency
N: 107, 339
Documentation as either a diagnosis, a set of ICD 9 codes or in laboratory results by serum creatinine [ 1.5 mg/dL
Teratogen exposure
Documentation of pre-specified teratogenic medications in active medication lista
Tobacco use
Documentation provided in social history or a set of ICD 9 codes
Sub-optimally controlled diabetes
Documentation in medical history, a set of ICD 9 codes or by Hgb A1c [ 6 % or laboratory result of serum glucose [200 mg/dL
Untreated Chlamydia
Positive laboratory results or set of ICD 9 codes without documentation of a corresponding macrolide or tetracycline in the medication list
Untreated gonorrhea
Positive laboratory results or a set of ICD 9 codes without documentation of a corresponding cephalosporin in medication list
Untreated syphilis
order to err on the side of over-identification rather than under-ascertainment. This study was performed according to prevailing ethical principles. We obtained institutional review board approval, and waivers of written consent and authorization were obtained to contact a subset of identified patients to participate in a survey. The survey instrument was designed to learn the status of the preconception health factors from the patients’ perspective without their knowledge of the algorithm results. Survey questions were initially reviewed for provider and patient face validity by obstetricians and a convenience sample of non-pregnant women. A random number was assigned to each of the potential survey subjects and patients numbered 1–1,000 were solicited to participate. If a patient did not respond within 1 month a reminder postcard was sent, and nonresponders at 2 months received a second mailed survey. Completed surveys triggered a detailed review of the corresponding patient EHR masked to the survey responses and the algorithm results. Comparisons were made between the algorithm, survey results and chart reviews for all health factors. Concordance was defined as the percentage of women who had agreement between their self-reported health factors by survey and the corresponding algorithm and manual chart
n: 29, 691
n: 1,000
n: 664
n: 336
n: 109
n: 83
Positive laboratory results or a set of ICD 9 codes without documentation of a corresponding penicillin or acceptable alternative in medication list
a
Teratogen medications categories included anticonvulsants, benzodiazepines, Coumadin/warfarin, 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors (statins), lithium, tetracyclines, and triptan medications
n: 141
Fig. 1 Flow diagram of surveys sent and analyzed
123
123
0 (0)
Sub-optimally controlled diabetes
0 (0)
2 (1.4) 1 (0.7)
4 (2.8)
12(8.5)
13 (9.2)
15 (10.6)
12 (8.5)
21 (14.9)
NA
83 (58.9)
106 (75.2)
0:141
0:139
0:140
7:133
12:110
1:136
3:130
22:113
12:118
32:79
141 (100)
139 (98.6)
140 (99.3)
140 (99.3)
122 (86.5)
137 (97.2)
133 (94.3)
135 (95.7)
130 (92.2)
111 (78.7)
Total n (%)
NA
0 (0)
0 (0)
0.93
0.48
-0.35
0.40
0.85
0.64
0.54
Kappac
0:140
0:137
1:129
7:127
8:110
1:128
3:116
NA
16:57
32:34
Pos:Neg n:n
140 (99.3)
137 (97.2)
130 (92.2)
134 (95.0)
118 (83.7)
129 (91.5)
119 (84.4)
NA
73 (51.8)
66 (46.8)
Total n (%)
0.50
0.49
0.52
0.71
0.11
0.48
0.42
NA
0.37
0.30
Kappa
Between algorithm and survey
0:140
2:137
0:129
7:128
5:112
3:128
7:120
NA
16:56
56:30
Pos:Neg n:n
140 (99.3)
139 (98.6)
129 (91.5)
135 (95.7)
117 (83.0)
131 (92.9)
127 (90.1)
NA
72 (51.1)
86 (61.0)
Total n (%)
0 (0)
0.66
0 (0)
0.68
0.20
0.27
0.46
NA
0.14
0.27
Kappa
Between survey and chart review
Pos positive, Neg negative, NA not asked
Kappa coefficients were calculated where j = 0 reflects an agreement purely due to chance and a j B 0.40 reflects a slight to fair agreement, and a j [ 0.40 reflecting a moderate to substantial agreement
c
b
Renal insufficiency, positive HIV status and untreated STD were not identified from the survey nor found in the chart abstraction or algorithm results Concordance defined as combined positive–positive and negative–negative health factor comparisons; listed in the first column as positive–positive:negative–negative and in the second column as total concordance (combined positive–positive and negative–negative)
a
0 (0)
ACE inhibitor use
0 (0)
1(0.7)
Anemia
2 (1.4)
7 (5.0)
1 (0.7)
Illegal drug use
7 (5.0)
8 (5.7)
7 (5.0)
Tobacco use
24 (17.0)
Hypertension
26 (18.4)
Obesity
18 (12.8)
19 (13.5)
17 (12.1)
Multivitamin use
60 (42.6)
Medical conditions/medications Teratogen exposure 24 (17.0)
33 (23.4)
Pos:Neg n:n
Between algorithm and chart review
Survey health factor
Algorithm health factor
Chart abstraction health factor
Concordanceb
Frequency n (%)
Alcohol use
Behavioral/nutritional
Health factor
Table 2 Frequency and concordance of preconception health factors derived from the algorithm, chart abstraction and patient surveya (n = 141)
Matern Child Health J
Matern Child Health J
review results. Chi squared analyses and Fisher’s exact tests were used to compare demographic and health factors between data gathering methods. Kappa coefficients were calculated where j = 0 reflects an agreement purely due to chance and a j B 0.40 reflects a slight to fair agreement, and a j [ 0.40 reflects a moderate to substantial agreement [19]. The Kappa coefficient is limited by prevalence, so positive–positive, negative–negative and total correlations were also reported. A p value of \0.05 was considered statistically significant.
Results Between 8/1/2010 and 2/29/2012 automated analysis of 107,339 outpatient encounters yielded 29,691 women who were of child-bearing potential. Patient ages were evenly distributed with 30 % between ages 18–24; 34 % between ages 25–35; and 36 % between ages 36–45. Over half (52.6 %) of these patients had C10 outpatient visits during the study period, suggesting that the health system and its providers were likely the primary source for their care. A majority of women had at least one preconception health factor (n = 19,624 or 66 %) while 8,652 (29 %) had C2 health factors and only 1,415 (5 %) had no health factors as assessed by the algorithm. Of the 1,000 patients randomly selected to receive a survey, 336 (33.6 %) were returned by the postal service (Fig. 1). After excluding wrong addresses with no forwarding address (n = 109), wrong addresses re-sent to forwarding address (n = 83) and 3 requested exclusions, 141 surveys were available for analysis. Preconception health factors identified via the algorithm, the chart abstraction and the patient surveys are noted in Table 2. Patients who returned the survey were similar in most respects to non-participants except that they were slightly older on average (32.7 ± 8.4 vs. 30.6 ± 8.3, p = 0.006) and more likely to have had increased contact with the health system during the study timeframe (C10 outpatient visits; 65.2 vs. 50.5 %, p = 0.001). The health factor profile derived from the algorithm among patients who returned the survey compared to those patients who did not was only different for one variable—alcohol consumption (23.4 vs. 12.0 %, respectively, p \ 0.001). The algorithm misidentified 43 women(30.5 %) as being of child-bearing potential by study definition when in fact, they were either using contraception (n = 39); already pregnant but receiving care outside of health system (n = 3); or not seeking pregnancy with their same-gendered partner (n = 1). Compared to the patient survey results, the algorithm and the chart review both underreported a majority of the preconception health factors secondary to incomplete EHR documentation (Table 2). The
algorithm results demonstrated high overall concordance (95.1 %) when compared to chart abstraction, suggesting that the case-finding logic accurately captured and interpreted the available discrete EHR entries made by providers. The next highest concordance was observed between patient survey results and chart abstraction (87.2 %), with similar concordance between survey results and the algorithm (85.8 %). Hypertension had the highest overall concordance between the algorithm and chart review (j = 0.93), the algorithm and survey (j = 0.71), and between the chart review and survey (j = 0.68). Where the algorithm performed less well compared to the survey (e.g., alcohol use (j = 0.27), the EHR chart appeared incomplete or incorrect.
Conclusions Our EHR algorithm identified almost 30,000 women seen as outpatients over 18 months, the majority of whom had potentially modifiable preconception health factors. When compared to patient reporting via our survey, the EHR chart and the algorithm underestimated health profiles but not by large margins. Missing or incomplete data entry in the electronic patient encounters was the primary explanation for this finding. Nonetheless, the algorithm performed well as indicated by its ‘‘internal validity’’ with 95.1 % average concordance compared to chart abstraction. The accuracy of the algorithm suggests that it could be useful in identifying at-risk women and flagging them for preconception counseling within the constraints of complete and accurate EHR documentation by providers. Both the algorithm and chart review appeared to underreport health factors when compared to the patient survey. Whether providers did not ask these questions, whether the answers were not recorded or if responses to selected health questions were misrepresented by patients is unclear. When Killeen and colleagues looked at concordance between self reported tobacco, alcohol and drug use compared to health record extraction they found discordance related to type of medical visit and insurance status [20]. Since all encounters queried were outpatient visits and insurance status was variable in our cohort, selective reporting by patients may have been at play. However, the under representation of the prevalence of alcohol and tobacco use by the algorithm and the chart review raises further concerns associated with incomplete data entry. Incomplete data entry can also have a potentially negative impact on patient safety well beyond our theoretic application in preconception health [21]. Missing data is also a common pitfall in EHR research [22] which we attempted to overcome by using multiple fields per factor to capture information entered in one of several possible
123
Matern Child Health J
locations. While we did not data mine free text entries, the efficiency of our algorithm suggests that adding this technology would provide only incremental improvement at best, while the potential downside would be significantly increased computer runtime (for the processing of 18 months of outpatient encounters our algorithm required approximately 24 h). Skeptics might fault the algorithm for misidentifying *30 % of the patients being of child-bearing potential owing to missing EHR data. This missing data rate is comparable to other studies which have reported ranges from *20–50 % [23–25] and we would assert that the algorithm is intended as a safety-net from which confirmation of health status would still be required as the next logical step before any patient-specific interventions would be undertaken. It appears that 29,691 women within our healthcare system were of child bearing potential with 95 % having at least one of the identified preconception health factors. Even if 30 % of these women were erroneously identified by our algorithm as being at-risk for pregnancy, this would still leave over 20,000 women who could potentially benefit from an intervention. We have no way of determining if our algorithm identified all women of child-bearing potential during this period (i.e., falsenegatives); however, we were conservative in the design of the algorithm to lean towards over ascertainment which represents a significant improvement above our baseline of usual care. The costs of over-identification would depend on the designed intervention which could range from an automated alert system in the EHR (inexpensive to deploy and maintain) to direct outreach by clinical personnel (requiring significant resources). Rakotz and colleagues designed an algorithm to screen for undiagnosed hypertension which identified 1,600 patients of which, 475 came in for confirmatory blood pressure measurements [14]. While they found that less than half (38 %) of patients flagged as hypertensive were confirmed to be truly hypertensive, the cost of their confirmatory clinical evaluations among normotensive cases must be considered in the context of potential health savings for hypertensive subjects who had not been identified previously and were atrisk for additional morbidity and cost without treatment. Patient recall bias could have impacted the accuracy of the survey results which in turn, could also have influenced concordance calculations with the algorithm and chart abstraction datasets. To try and limit this confounder, the interval between the actual patient encounters and the completion of the survey was within 12-months in a majority of subjects and the types of health questions posed were largely categorical. We also acknowledge that our case-finding algorithm remains vulnerable to errors in primary data entry. However, the 87.2 % average concordance rate comparing the survey to the chart abstraction
123
was at least comparable to best performance reported in another study of electronic data mining [23] and superior when compared to an observed accuracy rate of 30.7 % in a study reviewing automated data retrieval of 5,135 patient records for surgically related clinical questions. [26] In the future, EHR data accuracy may improve in response to meaningful use legislation linking complete discrete data field documentation to reimbursement [27]. Because comparative effectiveness research in preconception care is not yet available [28, 29] the proper strategy to modify risk for the women identified through our algorithm is unknown. Potential interventions could include communicating with the patients’ primary care providers via electronic physician alerts or computer-based decision tools, which have been shown to improve compliance with clinical protocols [30] and decrease teratogenic medication exposures in women without documented contraception [31]. When these electronic solutions are well designed, such innovations can have high acceptance rates from providers [32]. Alternatively, direct communication via a secure internet portal could invite the patient to be assessed for an undiagnosed or incompletely treated health condition or simply provide education on the importance of health management (e.g., vitamin supplementation). A pilot study looking at cardiovascular risk stratification has found that automated, personally targeted intervention strategies can work [16] and we have undertaken a similar study to improve preconception diabetes management. At present, over 200,000 patients use our EHR internet portal for results review, appointment scheduling and direct communication with their physicians. Since almost all (95.7 %) of the patients we surveyed reported being interested in receiving information about improving health prior to pregnancy, it would be reasonable to test this confidential web portal for such communications since it is already connected to the EHR system and could be an attractive platform for intervention. Further research is needed to determine whether preconception health improvements as reported by others [33] can actually be facilitated by first relying on an automated EHR screening algorithm similar to the one we have studied. Acknowledgments Thank you to Tom East, PhD, Doris Ng and Lourdes Link for their guidance on study design, algorithm development and data processing. This project was funded through intramural support. Conflict of interest None of the authors have a conflict of interest.
References 1. Centers for Disease Control and Prevention. (2006). Recommendations for improving preconception health and health care: United States: A report of the CC/ATSDR Preconception Care
Matern Child Health J
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Workgroup and the Select Panel on Preconception Care. MMWR Morbidity and Mortality Weekly Report, 55, 1–23. Kitzmiller, I. L., Buchanan, T. A., Kjos, S., Combs, C. A., & Ratner, R. (1996). Preconception care of diabetes, congenital malformations, and spontaneous abortions (technical review). Diabetes Care, 19, 514–541. Centers for Disease Control and Prevention. (1992). Recommendations for the use of folic acid to reduce the number of cases of spina bifida and other neural tube defects. MMWR Morbidity and Mortality Weekly Report, 41, 1–6. Honein, M. A., Paulozzi, L. J., Matthews, T. J., Erickson, J. D., & Wong, L. C. (2001). Impact of folic acid fortification of the US food supply on the occurrence of neural tube defects. JAMA, 285(23), 2981–2986. Rose, D. (2010) A history of the March of Dimes. March of Dimes Archives (Accessed 13 March, 2013. http://www.march ofdimes.com/mission/history_indepth.html). Olney, R. S., & Mulinare, J. (2002). Trends in neural tube defect prevalence, folic acid fortification, and vitamin supplement use. Seminars in Perinatology, 26(4), 277–285. Centers for Disease Control and Prevention. (1999). Knowledge and use of folic acid by women of childbearing age-United States, 1995 and 1998. MMWR Morbidity and Mortality Weekly Report, 48, 325–327. Finer, L. B., & Zolna, M. R. (2011). Unintended pregnancy in the United States: Incidence and disparities, 2006. Contraception, 84(5), 478–485. Taylor, A. K., Larson, S., & Correa-de-Araujo, R. (2006). Women’s healthcare utilization and expenditures. Women’s Health Issues, 16, 66–79. Morgan, M. A., Anderson, B. L., Lawrence, H., & Schulkin, J. (2012). Well-woman care among obstetrician-gynecologists: Opportunity for preconception care. Journal of Maternal-Fetal and Neonatal Medicine, 25(6), 595–599. Sankilampi, U., Saari, A., Laine, T., Miettinen, P. J., & Dunkel, L. (2013). Use of electronic health records for automated screening of growth disorders in primary care. JAMA, 310(10), 1071–1072. Anand, V., Carroll, A. E., & Downs, S. M. (2012). Automated primary care screening in pediatric waiting rooms. Pediatrics, 129(5), e1275–e1281. Seyfried, L., Hanauer, D. A., Nease, D., Albeiruti, R., Kavanagh, J., & Kales, H. C. (2009). Enhanced identification of eligibility for depression research using an electronic medical record search engine. International Journal Medical Informatics, 78(12), e13–e18. Rakotz, M. K., Ross, R. E., Robicsek, A., Konchak, C., Gavagan, T., & Ewigman, B. (2012). Searching for undiagnosed hypertension: A case study in the translation of quality research into operations. In International Health Care Quality Meeting, Paris, France, April 2012. Brenner, S., Detz, A., Lopez, A., Horton, C., & Sarkar, U. (2012). Signal and noise: applying a laboratory trigger tool to identify adverse drug events among primary care patients. BMJ Quality & Safety, 21(8), 670–675. Jones, J. B., Shah, N. R., Bruce, C. A., & Stewart, W. F. (2011). Meaningful use in practice using patient-specific risk in an electronic health record for shared decision making. American Journal of Preventive Medicine, 40(5 Suppl 2), S179–S186. Jack, B. W. Atrash, H., Coonrod, D. V., Moos, M. K., O’Donnell, J., & Johnson, K. (2008). The clinical content of preconception care: An overview and preparation of this supplement. American Journal of Obstetrics and Gynecology, 199(6 Suppl 2), S266–S279. Mansour, D., Inki, P., & Gemzell-Danielsson, K. (2010). Efficacy of contraceptive methods: A review of the literature. The
19. 20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
European Journal of Contraception and Reproductive Health Care, 15(1), 4–16. Viera, A. J., & Garrett, J. M. (2005). Understanding interobserver agreement: The kappa statistic. Family Medicine, 37(5), 360–363. Killeen, T. K., Brady, K. T., Gold, P. B., Tyson, C., & Simpson, K. N. (2004). Comparison of self-report versus agency records of service utilization in a community sample of individuals with alcohol use disorders. Drug and Alcohol Dependence, 73, 141–147. Institute of Medicine. (2001). Building organizational supports for change. In Institute of Medicine (Ed.), Crossing the quality chasm: A new health system for the 21st century (pp. 111–144). Washington, DC: National Academy Press. Terry, A. L., Chevendra, V., Thind, A., Stewart, M., Marshall, J. N., & Cejic, S. (2010). Using your electronic medical record for research: A primer for avoiding pitfalls. Family Practice, 27, 121–126. Prather, J. C., Lobach, D. F., Goodwin, L. K., Hales, J. W., Hage, M. L., & Hammond, W. E. (1997). Medical data mining: Knowledge discovery in a clinical data warehouse. In Proceedings of the AMIA annual fall symposium, p. 101–105. Botsis, T., Hartvigsen, G., Chen, F., Weng, C. (2010). Secondary use of EHR: data quality issue and informatics opportunities. In AMIA summits on translational science proceedings, p. 1–5. Newgard, C. D., Zive, D., Jui, J., Weathers, C., & Daya, M. (2012). Electronic versus manual data processing: Evaluating the use of electronic health records in out-of-hospital clinical research. Academic Emergency Medicine, 19(2), 217–227. Stein, H. D., Nadkarni, P., Erdos, J., & Miller, P. L. (2000). Exploring the degree of concordance of coded and textual data in answering clinical queries from a clinical data repository. Journal of the American Medical Informatics Association, 7(1), 42–54. Hasan, S., & Padman, R. (2006) Analyzing the effect of data quality on the accuracy of clinical decision support systems: A computer simulation approach. In AMIA annual symposium proceedings, p. 324–328. Zywiak, W., Metzer, J., & Mann, M. (2010). Meaningful use for eligible professionals: The top ten challenges. (Accessed March 13, 2013. http://www.csc.com/health_services/insights/44165-mean ingful_use_for_eligible_professionals_the_top_ten_challenges). Salihu, H. M., Salinas, A., & Mogos, M. (2013). The missing link in preconceptional care: The role of comparative effectiveness research. Maternal and Child Health Journal, 17(5), 776–782. Whitworth, M., & Dowswell, T. (2009). Routine pre-pregnancy health promotion for improving pregnancy outcomes. Cochrane Database of Systematic Reviews, 2009(4). doi:10.1002/14651858. CD007536.pub2. Garg, A. X., Adhikari, N. K., McDonald, H., Rosas-Arellano, M. P., Devereaux, P. J., Beyene, J., et al. (2005). Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: A systematic review. JAMA, 293(10), 1223–1238. Schwarz, E. B., Parisi, S. M., Handler, S. M., Koren, G., Cohen, E. D., Shevchik, G. J., et al. (2012). Clinical decision support to promote safe prescribing to women of reproductive age: A cluster-randomized trial. Journal of General Internal Medicine, 27(7), 831–838. Shannon, G. D., Alberg, C., Nacul, L., & Pashayan, N. (2013). Preconception healthcare and congenital disorders: systematic review of the effectiveness of preconception care programs in the prevention of congenital disorders. Maternal and Child Health Journal. doi:10.1007/s10995-013-1370-2.
123
