Reusing Electronic Patient Data for Dental Clinical Research: A Review of Current Status

INTRODUCTION

Traditional clinical research is considered to increasingly fall short of the needs of clinicians, patients and funding agencies for many reasons (1,2). They include the high cost of clinical trials, slow results, difficult enrollment, often poor generalizability, and challenges in discovering clinical outcomes and side effects in a nuanced and clinically meaningful manner (1,3,4).

In consequence, the reuse of electronic data collected during clinical care has received increased attention as a method for increasing our evidence base (3–7). Reusing data from electronic health records (EHR) is complementary to or synergistic with more traditional research approaches (3,8), and has a number of potential advantages. EHRs can support many types of research, ranging from epidemiology and outcomes of chronic diseases to pharmacovigilance, adverse drug events and comparative effectiveness (7). Using EHR data for research can help increase efficiency (1,4), lower research costs (9), allow the study of patient rather than research participant populations, avoid certain selection biases, implement longitudinal studies, detect rare events, and discover drug side effects earlier than possible with traditional methods (10–12).

Reuse of electronic dental record (EDR) data has become increasingly attractive in the context of the National Dental Practice-based Research Network (NDPRN) initiative by the National Institute of Dental and Craniofacial Research (NIDCR). The experience during the practice-based research network (PBRN)’s initial funding period has shown that there are more relevant research questions than can be practically addressed using the typical PBRN study approach; that some important long-term research questions do not fit very well into the timeframe of a PBRN study; and that costs and practice workflow issues limit the “throughput” of the PBRN system. Given the fact that approximately 75% of DPBRN practitioners use a computer to manage clinical information and 15% are paperless (13), secondary analysis of EDR data is increasingly compelling.

Figure 1 illustrates the workflow of a continuous cycle of improvement based on the analysis of electronic patient data in the context of a Learning Health Care System (3). Initially, patient data generated during the clinical encounter are captured electronically, ideally in standardized form. A data extraction, validation and analysis process produces answers to clinical questions. Once disseminated, those answers can help change clinical practice, resulting in improved care outcomes.

Figure 1.

Workflow of the Learning Healthcare System, in which reuse and analysis of electronic patient data forms the basis of a continuous quality improvement process

Clearly, data from EHRs have multiple limitations compared to the data collected in well-designed and –executed clinical trials. First, EHR data are collected for clinical, not research purposes (6,14). Resulting biases can range from threats to the representativeness of the population and clinician-related biases to missing data and poor characterization of outcomes (3). Second, data in EHRs tend not to be very standardized for multiple reasons (6,15). Many medical systems still favor free text over structured data entry (16); allow users to store the same or similar information in multiple places; and validate data inconsistently. Last, EHR data exhibit variable levels of accuracy (17,18) and typically provide poor support for systematic data extraction (6).

Despite these limitations, EHRs are increasingly seen as a potential source of data for research. This is evidenced in three large-scale initiatives in medicine using electronic medical records (EMR): the Distributed Ambulatory Research in Therapeutics Network (DARTNet) Institute (19), the electronic Primary Care Research Network (ePCRN) Consortium (20), and the Deliver Primary Health Care Information (DELPHI) project in Canada (14). In dentistry, the Consortium for Oral Health Research and Informatics (COHRI) (21) has created a data warehouse for research from patient records at several dental schools (see Appendix A).

As this review shows, there are three compelling reasons to develop our capabilities for reuse of EDR data collected during patient care. First, reuse of EHR data for research has added significantly to our capacity to generate knowledge from routine clinical care. Second, EDR data are a potentially valuable data source for clinical, comparative effectiveness, epidemiology and other research, especially given NIDCR’s recent emphasis on practice-based research. Last, electronic data are increasingly available from private practices due to the rapid adoption of computers. We therefore performed a review of the current status of reuse of electronic patient data for dental research, guided by the following questions: What types of research projects have used EDR/EMR data and what were the characteristics of these studies? What research questions did they examine? How did study characteristics relate to data sources, either EMR, EDR or both? What study variables were extracted from EDR and EMR systems? What advantages did reusing EDR/EMR data convey to these studies? What barriers/limitations were reported?

MATERIALS & METHODS

Data Sources and Search strategy

A list of MeSH terms, Emtree (Embase) thesaurus terms and general keywords was developed to search the MEDLINE and Embase databases. A biomedical librarian (RA) tested multiple combinations of terms to optimize the search (see Appendix B). Searches were limited to English, French, German, Finnish, Norwegian and Swedish language articles between January 1992 and January 2013. The database search was supplemented with hand-searching seven main journals that publish dental and informatics research: Journal of American Dental Association, Journal of Public Health Dentistry, Journal of Periodontology, Journal of Dental Research, Journal of Clinical Periodontology, Journal of Dentistry and Journal of the American Medical Informatics Association. Authors of 35 relevant papers were contacted to recommend additional studies. Searches were also conducted in the Publication Archive of the Dental Informatics Online Community¹, an archive of ~ 2,300 articles on informatics and information technology in dentistry.

Study Selection Criteria and Process

Studies on dental and craniofacial topics, alone or in combination with medical aspects (such as conditions, medications and outcomes), were included. Studies had to include at least some data recorded electronically during routine patient care, using electronic records in the clinic. Studies were excluded if they exclusively analyzed electronic insurance/claim, administrative database, patient registry and hospital discharge data; databases that derived data from EDRs/EMRs; electronic data converted from paper records; research registries; and national survey databases. Systematic reviews were also excluded.

Three reviewers (MS, KL and TS) independently screened titles and abstracts of studies retrieved by the initial search for preliminary inclusion. Disagreements were resolved through discussion. Full-text articles for all candidate studies were then reviewed for final inclusion. When information on a study was unclear, the study authors were contacted for clarification.

Data Extraction

The data collected about each study are shown in Table 1. Three reviewers extracted qualitative and quantitative data from the studies independently.

Table 1.

List of data extracted from the studies.

Type of research*	epidemiology: prevalence, incidence and risk factors of dental and/or medical diseases/conditions health services: access to, delivery and quality of dental and medical services outcomes: patient outcomes of dental and medical treatments other: studies that did not fall into the previous three categories
Study design	cross-sectional, case control, cohort, longitudinal, quasi-experimental, randomized clinical trials
Type(s) of data source(s)	EDR, EMR, paper records, film radiographs, surveys
Study population	adults, children, soldiers/veterans
Research questions	e.g. survival of root canal therapy
Study variables	e.g. patient-level variables, treatment variables, medications
Publication year	e.g. 1992, 2012
Sample size	<1000, 1000–5000, 5001–99,999, ≥100,000
Follow-up period	<3 years, 3–9 years, ≥10 years
Advantages and limitations of using EDR/EMR data	extracted and summarized from the articles

^*While different definitions exist for the types of research, we used the listed definitions for our study.

Data Analysis

Data analysis followed the guidelines by Popay et al. in 2006 (22). Though this study is not a systematic review, the approach was selected because it provides a coherent framework and permits the synthesis of studies that explore a wide range of research questions with various research designs (23). First, the authors developed a preliminary synthesis by grouping the studies by the type of research they belong to. Data were tabulated and compared to see if type of research correlates to EDR/EMR data usage in certain ways. Then, the studies were regrouped based on the research questions they addressed. Thematic analysis was conducted to identify patterns on how EDR/EMR data was used in relation to specific study characteristics. For instance, does research question dictates the use of certain data source? Whenever possible, we also conducted descriptive quantitative analysis. The first author conducted the initial analysis, checked with the research team and conducted additional analysis based on feedback. The team discussed and reached iteratively consensus about the analysis results through multiple meetings.

RESULTS

Study Selection

Figure 2 summarizes the results of search and review. Of an initial 1,527 studies, 98 met the criteria for preliminary inclusion, of which 60 remained after detailed review.

Figure 2.

A flowchart of the process of study selection

Study Characteristics

Table 2 summarizes the key characteristics of the studies. More detailed information is provided in Table 3. Figure 3 provides an additional representation that relates research topics to the type of electronic data used.

Table 2.

A summary of key characteristics of studies in the review (n = 60)

Study characteristics	Frequencies	Percentages
Type of research
Epidemiology	32	53
Outcomes	16	27
Health services	10	17
Other	2	3
Research design
Cross-sectional	25	42
Cohort	20	33
Case control	8	13
Longitudinal	5	8
Quasi-experimental	1	2
Randomized control trial	1	2
Sample size
1 – 999	35	57
1000 – 5000	11	19
5001 – 99,999	11	19
100,000 +	3	5
Follow-up period
0 – 2 years	26	43
3 – 9 years	23	38
10 + years	11	19
Patient care setting
Dental school clinics	15	25
Dental HMOs	11	19
VA medical centers	11	19
Public dental health centers/nursing home	10	17
Other hospitals/clinics	7	10
Private dental practices	6	10
Research setting
VA medical centers	9	15
Kaiser Permanente	9	15
Dental insurance groups	3	5
Other	39	65

Table 3.

A list of all studies included in the review

#	Author (yr)	Study domain	Study design	Research questions	Country	Setting	Patient population
Outcomes research
1	Hujoel, Leroux, Selipsky, and White (2000)	Dental treatment & medical condition	Case control	Association between tooth loss and non- surgical periodontal treatment	US	Dental HMOs	Adults
2	Aquilino, Shugars, Bader, and White (2001)	General treatment	Longitudinal	Survival of teeth adjacent to treated and untreated missing tooth	US	Dental HMOs	Adults
3	Stewart, Wager, Friedlander, and Zadeh (2001)	Dental treatment & medical condition	Case control	Effect of periodontal therapy on glycemic control	US	VA medical centers	Veterans
4	Aquilino and Caplan (2002)	Endodontic treatment	Cohort	Association between crown placement and survival of endodontically treated teeth	US	Dental school clinic	Adults
5	Caplan, Kolker, Rivera, and Walton (2002)	Endodontic treatment	Cohort	Relationship between number of proximal contacts and survival of RCT teeth	US	Dental school clinic	Adults
6	Jones et al. (2007)	Dental treatment & medical condition	Randomized clinical trial	Efficacy of periodontal care on glycemic control	US	VA medical centers	Veterans
7	Caplan, Cai, Yin, and White (2005)	Endodontic treatment	Cohort	Impact of pulpal involvement and RCT on tooth survival	US	Dental HMOs	Adults
8	Shelley, Johnson, and BeGole (2007)	Restoration & endodontic treatment	Cross-sectional	Use of EPR for quality of care assessment	US	Dental School clinic	Adults
9	Leskinen et al (2008)	Dental prevention	Cohort	Outcome of sealant treatment of first	Finland	Public health	Children
10	Leskinen et al (2008)	Dental prevention	Cohort	Cost-effectiveness of two caries permanent preventive strategies	Finland	Public health centers	Children
11	Bassim, Gibson, Ward, Paphides, and Denucci (2008)	Dental prevention & medical condition	Quasi- experimental	Association between use of oral hygiene aids and pneumonia mortality	US	Nursing homes	Adults
12	Iqbal, Kurtz, and Kohli (2009)	Endodontic treatment	Cross-sectional	Incidence and factors to endodontic flare- ups	US	Dental school clinic	Adults
13	George et al. (2011)	Implant treatment	Longitudinal	Clinical outcomes of a new implant Ti- Unite	US	Public hospital	Adults
14	Worley et al. (2012)	General treatment	Cohort	Outcomes of restoration placed by dentists and RFAs	US	Dental HMOs	Adults
15	Helenius-Hietala, Aberg, Meurman, and Isoniemi (2012)	Dental treatment & medical condition	Longitudinal	Dental treatment and post liver transplant infection risk	Finland	University hospital	Adults
16	Kolhatkar et al (2012)	Dental treatment & medical condition	Cohort	Outcome of CLS in HIV patients	US	Dental school clinic	Adults
Health services research
17	Ojima, Hanioka, and Shizukuishi (2001)	Periodontal treatment	Cohort	Risk factors to compliance of SPT	Japan	Dental school clinic	Adults
18	Ladrillo, Hobdell, and Caviness (2006)	Dental condition	Cross-sectional	Prevalence and characteristics of pediatric ED for dental problems	US	Children’s hospital	Children
19	Okunseri, Bajorunaite, Matthew, and Iacopino (2007)	General treatment	Cross-sectional	Racial/ethnic differences in the provision of dental procedures	US	Dental school clinic	Adults
20	Luzzi and Spencer (2009)	General treatment	Longitudinal	Public dental service utilization	Australia	Public dental service	Adults
21	Okunseri, Bajorunaite, Mehta, Hodgson, and Iacopino (2009)	Dental prevention	Cross-sectional	Gender difference in receipts of preventive dental treatment procedures	US	Dental school clinic	Adults
22	Naegele, Cunha- Cruz, and Nadanovsky (2010)	General treatment	Cohort	Disparity between dental treatment needs and actual treatment completed	Brazil	Dental private practices	Adults
23	Eikenberg, Keeler, and Green (2011)	General treatment	Cross-sectional	Dental care to soldiers in Iraq	US	Military clinics	Soldiers
24	Kudiyirickal and Hollinshead (2011)	General treatment	Cross-sectional	Prescription of antimicrobials for orofacial infections by dentists	UK	Dental private practices	Adults
25	Chaudhari et al (2012)	General treatment	Cross-sectional	Dental care utilization between adults with/without diabetes	US	Various dental & medical providers	Adults
26	Morgan et al. (2012)	Dental condition	Cross-sectional	Oral health status of people with intellectual and developmental disabilities	US	Dental school clinic	Adults
Epidemiology
27	Caplan and Weintraub (1997)	Endodontic treatment	Case control	Risk factors to loss of RCT teeth	US	Dental HMOs	Adults
28	Caplan and White (2001)	Endodontic treatment	Case control	Risk factors to noncompletion of RCT	US	Dental HMOs	Adults
29	Elter, White, Gaynes, and Bader (2002)	Dental & medical condition	Cohort	Association between depression and periodontal treatment outcome	US	Dental HMOs	Adults
30	Fouad and Burleson (2003)	Dental treatment & medical condition	Cohort	Diabetes and endodontic treatment outcome	US	Dental school clinic	Adults
31	Lee and Dennison (2004)	Dental prevention	Cross-sectional	Fluoridated water and oral health difference in children	New Zealand	Hospital/public health services	Children
32	Byrappagari, Mascarenhas, and Chaffin (2006)	Dental condition	Cross-sectional	Association between DFC and caries and tobacco risk assessment	US	Military clinics	Soldiers
33	Maupomé, Peters, Rush, Rindal, and White (2006)	Dental condition & medication	Cohort	Association between Xerogenic medications and crown/root restorations	US	Dental HMOs	Adults
34	Meurman, Furuholm, Kaaja, Rintamaki, and Tikkanen (2006)	Dental & medical condition	Cross-sectional	Association between oral health and pregnancy outcome	Finland	Maternity health centers	Adults
35	Hyde, Bader, and Shugars (2007)	Endodontic treatment	Cross-sectional	Incidence of and risk factors of provisional crown failures	US	Dental school clinic	Adults
36	Korhonen, Salo, Suni, and Larmas (2007)	Dental condition	Cohort	Oral health differences in Finland	Finland	Public health centers	Adults & children
37	Forsberg, Sjodin, Lundgren, and Wanman (2008)	Dental condition	Case control	Use of EDR in monitoring dental health	Sweden	Public dental services	Adults
38	Friedlander, Tajima, Kawakami, Wang, and Tomlinson (2008)	Dental & medical condition	Case control	Relationship between nutritional status and masticatory function	US	VA medical centers	Veterans
39	Tavares et al. (2008)	Dental condition	Case control	Protection factors related to dental caries in children	Brazil	Public oral health service	Children
40	Korhonen, Gundagar, Suni Salo, and Larmas (2009)	Dental condition	Cohort	DMF index differences in new and old patients	Finland	Public health centers	Adults
41	Ram, Teruel, Kumar, and Clark (2009)	Dental condition	Cross-sectional	Prevalence and characteristics of patients with atypical odontalgia	US	Dental school clinic	Adults
42	Sedghizadeh et al (2009)	Dental condition & medication	Cross-sectional	Association between oral bisphosphonate and ONJ	US	Dental school clinic	Adults
43	Dissick et al. (2010)	Dental & medical condition	Cohort	Association between periodontitis and rheumatoid arthritis	US	VA medical centers	Veterans
44	Friedlander, Sung, Chung, and Garrett (2010)	Dental & medical condition	Case control	Relationship between radiographic atheromas and chronic dental infection	US	VA medical centers	Veterans
45	Kidd, Beattie, and Campbell-Hewson (2010)	Dental condition	Cross-sectional	Demographics and mechanisms of facial injury in children	UK	Pediatric ED	Children
46	Sanders and Slade (2010)	Dental & medical condition	Cohort	Association between dental caries risk and Apgar score	Australia	Public school dental clinics	Children
47	Spangler et al (2010)	Dental treatment & medical condition	Cross-sectional	Association between periodontal diseases and GHb level	US	Various dental & medical providers	Adults
48	Alman et al. (2011)	Dental treatment & medical condition	Cohort	Association between periodontal diseases and CHD	US	VA medical centers	Veterans
49	Fellows et al. (2011)	Dental condition & medication	Cohort	Oral bisphosphonate and incidence of ONJ	US	Dental HMOs	Adults
50	Kassab et al. (2011)	Dental treatment & medical condition	Cross-sectional	Treatment for heart diseases and invasive dental care	US	Dental school clinic	Adults
51	Mador et al. (2011)	Dental & medical condition	Cross-sectional	Association of OSA and risk of cardiorespiratory complications	US	VA medical centers	Veterans
52	Tami-Maury et al (2011)	Dental condition & medical treatment	Cohort	Risk factors of oral manifestation among HIV patients	US	University clinic	Adults
53	Newton et al. (2011)	Dental & medical condition	Cross-sectional	Prevalence of periodontal care among patients with/without diabetes	US	Various dental & medical providers	Adults
54	Chang, Tanner, Harada, Garrett, and Friedlander (2012)	Dental & medical condition	Cross-sectional	Prevalence of atheromas of patients with OSA vs patients with syndrome Z	US	VA medical centers	Veterans
55	Kudiyirickal and Hollinshead (2012)	Dental condition	Cross-sectional	Characteristics of orofacial infections in dental patients	UK	Dental private practices	Adults
56	Mosen, Pihlstrom, Snyder, and Shuster (2012)	Dental treatment & medical condition	Case control	Receipt of dental care on diabetes control and care utilization	US	Dental HMOs	Adult
57	Tanner et al. (2012)	Dental & medical condition	Cross-sectional	Association between OSA and Metabolic Syndrome	US	VA medical centers	Veterans
58	Yamazaki et al (2012)	Dental condition & medication	Cohort	Risk of oral BPs and other drugs for ONJ	Japan	University hospital	Adults
Other
59	Powell, Leroux, Martin, and White (2000)	Dental condition	Cross-sectional	Use of insurance claim data in identifying caries	US	Dental HMOs	Adults
60	Morgan et al. (2012)	Dental prevention	Cross-sectional	Use of EDR to track tobacco use and monitor cessation interventions	US	Dental school clinic	Adults

Note: Abbreviations used in this table are: RCT – Randomized control trial; EPR – Electronic patient records; RFAs – Restorative function auxiliaries; CLS – Crown lengthening surgery; SPT – Supportive periodontal therapy; ED – Emergency department; DFC – Dental fitness classification; EDR – Electronic dental records; DMF – Decayed, missing and filled teeth; ONJ - Osteonecrosis of the jaw; CHD – Coronary heart disease; OSA – Obstructive sleep apnea; BPs - Oral bisphosphonates

Figure 3.

Sample size, publication year and follow-up period for all studies arranged by research topic and use of electronic data source

1. What types of research projects have used EDR/EMR data and what were their characteristics?

In summary, more than half of the studies (32 studies, 53%) addressed research questions on epidemiological topics. All but two studies were retrospective, with cross-sectional as the leading study design (25 studies, 42%). The majority of studies (43, 72%) were conducted in the US. Follow-up periods ranged from 1 to 12 years, with patient data from as early as 1985 and as recent as 2011. Sample sizes varied substantially, from 9 patients with osteonecrosis of the jaw (ONJ) (24) to 153,619 general patients (25). Fifty-three studies focused on adult patients and 7 on children. Patients from 15 studies (25%) were seen at dental school clinics, with dental care provided by private practitioners in only 6 studies. In research settings, researchers from three organizations conducted one-third of the studies (21 studies). Thirty-five studies (58%) used exclusively electronic data while the remaining 25 studies were supplemented with paper data.

2. What research questions did these studies examine?

The studies addressed a variety of research questions. The majority of epidemiological studies investigated the association of risk factors with various dental or medical conditions, such as fluoridated water and caries (26), bisphosphonate therapy and ONJ (24,27), prevalence of atheromas and Obstructive Sleep Apnea (OSA) (28), and periodontal disease and depression (29). Ten health services studies examined access to care, such as pediatric emergency visits for dental problems (30), public dental service utilization (31), and care delivery, such as dental care to soldiers (32), racial difference in provision of dental procedures (33), and disparities between treatment needs and provision (34). Sample outcomes studies looked at the survival of root canal filled teeth (35–37), the effect of periodontal treatment on diabetes control (38) and endodontic treatment outcomes (39).

3. How did study characteristics relate to data sources, either EDR, EMR or both?

When the type of research was correlated with the use of EDR/EMR data, no pattern emerged for either epidemiological or outcomes studies. For health services studies, however, 9 out of 10 studies used exclusively electronic data with one supplementing with paper patient registration data (40).

The relationship between research questions/topics and use of data source was then examined (Figure 3). The 12 studies that described the prevalence and characteristics of a dental disease or a population’s oral health status tended to use only EDR (8 studies) or EMR (4 studies) data. They were often cross-sectional studies with a less than 5-year follow-up period. For instance, Ladrillo et al. in 2006 analyzed two hospital electronic databases to describe the characteristics of pediatric ER visits for dental problems (30), and Ram et al.’s 2009 study used an EMR to determine the prevalence and characteristics of patients with atypical odontalgia (AO) (41).

Among studies on dental treatments, 10 looked at general dental services utilization, of which 9 used EDR data only. Seven studies investigated the risk factors/outcomes of endodontic procedures. All except one relied on both EDR and other data sources and followed patients for 6 years or more. As these studies attempted to identify various risk factors, they required variables at the patient, treatment and tooth levels, some of which may be unavailable in EDRs. In Caplan et al.’s 2002 study on the relationship between the number of proximal contacts and survival of root canal therapy, patient demographics and provider were extracted from an EDR; tooth arch, type and treatment complications from paper treatment notes; and radiographs and dental/medical histories from health questionnaires (37).

For studies exploring the impact of medications on dental diseases, use of data sources depended partially on where medications were documented. In four studies, one used an EDR exclusively, one both EDR and EMR, and two EMR data. Even for a similar research question, data sources varied. For example, to study the relationship between bisphosphonate use and ONJ, Sedghizadeh et al. queried an AxiUm database in 2009 for ONJ diagnosis variables (24) while Fellows et al. resorted to an EMR for the same data in their 2011 study (27).

Another group of 24 studies examined oral-systemic connections, among which 10 explored the association between dental and medical conditions and an equal number the effect of dental treatment on medical conditions. Given the dual focus, these studies used both dental and medical data; however, the two subgroups revealed different patterns in electronic data usage. In the first group, 8 of 10 studies used VA’s integrated electronic medical/dental record to study topics such as periodontal disease and coronary heart disease (CHD) (42) and chronic dental infection and atheromas (43), with only 2 studies using other data sources. Conversely, for studies on dental treatment effect, 7 used data other than electronic sources and, more notably, all medical data were from EMR but dental data from other sources. For example, in Jones et al.’s 2007 study on the efficacy of periodontal care on glycemic control, HbA1c values were extracted from medical records but all periodontal treatment data acquired through a clinical examination (44).

4. What study variables were extracted from EDR and EMR systems?

Table 4 lists the variables most commonly used in the studies and from what sources they were extracted. Predominant patient-level variables were age, gender, type of insurance, but rarely socioeconomic status. Diagnostic variables referred to dental and medical diagnoses, such as caries and oral infections. Treatment procedures, particularly restorative, endodontic and periodontal treatments, were common. In studies involving medications, medication name, dosage and duration were often used. The most extracted medical history variables were diabetes diagnosis, glucose level and smoking status/use. As the results show, patient demographics and medication variables may be available in both EMRs and EDRs; treatment and tooth variables mostly in EDRs; and medical history in EMRs. Variables extracted the least from EMRs/EDRs were health behaviors and social factors.

Table 4.

Study variables extracted from electronic records and other sources

Study variables extracted	Total # of studies	# with EDR data	# with EMR data	# with other sources
Patient-level variables
Age	43	22	14	7
Sex	34	17	10	7
Race/ethnicity	16	4	9	3
Weight/height/BMI	15	0	14	1
Insurance	13	7	6	0
Social-economic status	11	3	3	2
Vital status	3	0	3	0
Medical/primary care use	3	0	3	0
Marital status	3	2	1	0
Education level	2	0	1	1
Other patient-level variables (e.g., birth   weight, vaccination, Apgar scores, etc)	20	6	8	6
Dental treatment variables
Restorative treatment	15	13	0	2
Endodontic treatment	14	11	0	4
Periodontal treatment	13	6	0	7
Surgical treatment	9	7	0	2
PDTP treatment	6	6	0	0
Surgeries (ENT, face, etc)	3	3	0	0
Orthodontic treatment	3	1	0	2
Dentures	3	3	0	0
Other treatment variables (e.g.,   diagnoses, techniques, codes, etc)	15	13	1	1
Tooth-level variables
Caries activities	13	8	1	4
Dmfs and/or DMFS	8	8	0	0
Tooth type	7	5	0	2
# of proximal contacts	4	0	0	4
Tooth arch	3	0	0	3
Periapical lesion on root	3	0	1	2
Tooth number	2	2	0	0
Pulp diagnosis	2	0	0	2
Other tooth-level variables (# of good,   missing teeth, alveolar bone loss, etc)	17	5	0	12
Dental history
Gums bleeding	6	0	0	6
Orofacial injuries/infections	4	1	0	3
Osteonecrosis of the jaw	3	2	1	0
Obstructive sleep apnea	3	0	3	0
Dental anxiety	2	0	0	2
History of wearing denture	2	0	0	2
Atypical odontalgia	1	1	0	0
Other dental conditions (e.g. pain, dry   socket, etc)	9	6	2	1
Medical history
Diabetes (diagnosis, glucose level,   GHb, A1c)	17	3	12	2
Smoking status/use	16	5	10	1
Heart diseases	4	2	2	0
Immune disorders (HIV, lupus, etc)	3	0	1	2
Cancer	3	1	2	0
Hypertension	3	1	2	0
Osteoporosis	3	1	2	0
Cholesterol values (HDL, LDL,   triglycerides)	3	0	3	0
Other medical conditions (e.g.   dementia, pneumonia, injuries, etc)	20	4	10	6
Medications
Name/type of medication	21	7	9	5
Numbers of medication	4	0	1	3
Dosage/duration of medication	4	2	2	0
Health behaviors/risks & social factors
Oral status (brushing, flossing, stain)	5	0	2	3
Fluoridated water	4	3	0	1
Alcohol/substance use	4	0	3	1
Sugar consumption	3	0	0	3
Mental health	2	1	1	0
Caries/perio risk	4	4	0	0
Oral health (good, fair, etc)	2	0	0	2
Sore risk	1	0	1	0
Fall risk	1	0	1	0
Grind or clench teeth	1	0	0	1
Activity level	1	0	0	1
Participation in community	1	0	1	0
Other variables
Provider (type, age, sex)	9	9	0	0
Number of dental/medical care visits	6	4	0	2
Date of visits	5	4	1	0
Experience of provider	2	2	0	0
New vs. old patients	2	2	0	0
Type of visits	2	2	0	0
Home-hospital distance	1	1	0	0
Diabetes-related visits/hospital   admission	4	0	4	0

5. What advantages did reusing EDR/EMR data convey to these studies?

Forty-nine of the 60 studies reported strengths and/or limitations of using electronic data, typically as part of the study limitations (Table 5). The most frequently reported advantage was the ability to conduct studies of significant statistical power due to large sample sizes. For instance, in Newton et al.’s 2011 study on periodontal care and diabetes, linking electronic dental claims and medical record data yielded a sample of more than 46,000 people with data about medical and dental care, one of the largest of its type in the literature. Access to data on multiple medical conditions allowed researchers to obtain well-matched controls and ascertain important potential confounders (45). Similarly, Korhonen et al. in 2009 were able to compare the decayed, missing and filled teeth (DMF) index of 153,619 patients in two Finish towns through access to digital data in public dental offices (25).

Table 5.

Advantages and limitations of using EMR/EDR data reported in the studies

Advantages & limitations of using EMR/EDR	Studies
Advantages
• Study diseases/conditions in a large sample with more statistical power	(14), (31), (42), (47), (53), (60)
• Save time compared to reviewing paper charts/records	(14), (36), (59)
• Provide a rich resource for outcomes research	(29), (35), (58)
• Identify patients with rare diseases (ONJ, AO, etc)	(41), (49)
• Track patients for an extended period of time	(30), (31), (60)
• Provide valid and representative sample	(23), (26)
• More systematic data documentation	(15), (51)
• Flexible to add specific fields for data collection	(30)
• More valid data than self-reported data	(20)
• Provide data on comorbid medical conditions	(6)
• Ability to collect data in real time	(52)
• Ability to ascertain potential confounders	(53)
Limitations
• Some study variables not documented in EDR (history of tobacco use, socio-economic status, education, compliance with treatment, etc)	(1), (2), (4), (5), (7), (8), (19), (23), (25), (28), (29), (37), (49), (53), (54), (56)
• Inaccurate, incomplete or missing data	(8), (15), (19), (27), (29), (36), (45), (46), (49), (50), (55), (58), (60)
• Study sample limited to patients of one EDR/institution	(18), (21), (37), (41), (57)
• Study sample limited to only male population	(6), (43), (44), (54), (57)
• Inconsistency in data documentation	(34), (36), (40), (46), (53)
• Data validity and reliability in question	(2), (4), (7), (31), (52)
• Inability to control experimental conditions	(60)

Electronic data were also valuable in providing a rich resource for outcomes research, identifying patients with rare diseases and saving study time. Longitudinal clinical data available in both an EMR and EDR facilitated a rich study on the effect of periodontal treatment outcome on depression (29), failures of provisional crowns (46) and the relationship of osteoporosis medications and ONJ (47). For identifying relatively uncommon diseases, such as OA or ONJ, Ram et al’s 2009 study (41) used EMR search functions to identify 64 OA patients in a population of 3,000 over 5 years. Querying an EDR and an EMR, Sedghizadeh in 2009 identified 9 out of 208 patients on alendronate with a diagnosis of active ONJ (24). Other less frequently mentioned advantages of using electronic data included the ability to collect data in real time (48) and more systematic data documentation (49).

6. What barriers or limitations of reusing EDR/EMR data were reported?

In terms of limitations, over half of the studies considered data availability and quality as major problems. Specifically, 16 studies were unable to obtain study-specific variables not routinely captured in EDRs; 12 reported inaccurate, incomplete or missing data; and 5 cited inconsistency in data documentation as a limitation. For instance, socioeconomic status, race, education, smoking and treatment compliance were often used as confounding variables, but they were not routinely documented in most EDRs. In the studies on root canal treatment, variables such as degree of bone loss and number of proximal contacts, were not available electronically (37,39). In other studies, data on patient visits to other providers and pharmacy records on medication usage were not accessible (42,50).

A second limitation was inaccurate, incomplete and missing data, potentially attributed to coding errors and inconsistent data documentation practices when multiple providers or uncalibrated examiners entered data. For example, the recording of the DMF index between dentists varied greatly in Finland, resulting in discrepancies in the data collected from various dental offices (25). In Sanders et al’s 2010 study on the relationship between Apgar score and dental caries risk for children, missing Apgar scores as a primary variable for some participants posed a challenge for statistical analysis (51).

In addition, characteristics of study populations sometimes raised concerns about selection bias and reduced the generalizability of study results, such as in the VA studies with their patient population of predominantly male veterans (52,53).

DISCUSSION

This review has yielded important early insights into the current state of the reuse of EDR/EMR data for dental clinical research. The fact that we were able to identify only 60 studies published between 1992 and 2012 (40% of them published since 2010) indicates that the reuse of electronic data in dentistry is in an early stage but rapidly accelerating. Evidence suggests that the number of medical studies reusing electronic patient data is significantly larger. Key findings from this review include:

Reuse of electronic data supports primarily retrospective research

Electronic patient data are relatively easy and cheap to reuse for research. However, they can only be “after the fact,” restricting most studies to analyzing data retrospectively. However, within this constraint, a variety of study designs are feasible, such as cross-sectional, longitudinal and cohort studies. Even a prospective approach is possible, as long as the necessary variables are accommodated ahead of the study and captured in the system by clinicians (38,44).

Research is focused on epidemiology and risk factors

The fact that thirty-two of 60 studies addressed questions related to epidemiology and risk factors is likely due to the ease with which data about conditions such as caries and periodontal disease can be extracted from highly structured EDRs. Health services research studies (10 in our review) may benefit from the fact that dental records contain more clinical variables than insurance claims databases. Our review did not locate comparative effectiveness research studies, for which electronic patient records are considered to be an essential platform (54).

Research combining EDR and EMR data is limited

Only 10 of the 24 studies that explored oral-systemic connections used both EDR and EMR data, while the remainder supplemented electronic data with clinical exams, questionnaires and insurance claims. Research combining electronic dental and medical data is difficult unless they are maintained in a way that allows them to be linked easily, such as at the VA and HealthPartners.

EDRs in private practice are essentially untapped

Only three studies analyzed EDR data from private dentists, suggesting a huge gap in addressing clinical research questions in private practice. We may be able to reduce this gap as more dentists adopt EDRs, as suggested by our recent survey (13).

EDRs/EMRs often need to be supplemented with other data sources

Close to half the studies supplemented EDR/EMR data with data from other sources. Our results (Table 4) provide important guidance on which data can most likely be found in which data source(s).

Contextual and cultural factors determine study feasibility

Several studies in Europe took advantage of standardized EDRs and databases for monitoring oral health at the local and national level. A higher degree of standardization of electronic records and better adoption of public health databases and registries make data more easily available to dental research than in the US.

While many of the studies appear to have been successful in addressing important research questions, they also highlighted important current limitations in reusing electronic data:

Study samples are often idiosyncratic and not generalizable to the population at large

While the studies in this review often used samples much larger than typically in traditional studies, many were constrained by the population covered in the database(s). For instance, most studies included only patients, not people, limiting inferences for the general population. Depending on the utilization of dental services (in the US, about 70% of the population have at least one dental visit per year, and in Europe the number ranges from 30%–85% across countries) (http://www.cecdo.org/pages/database intro.html), such studies are bound to exclude large segments of the population. Others projects studied special populations, such as military veterans and patients enrolled in a health maintenance organization (HMO), also reducing their generalizability.

Methodological issues threaten validity

Studies reusing data collected in the course of clinical practice differ from rigorously conducted clinical studies in many ways. Clinicians are typically not as well calibrated and trained as examiners in a research study. Inter-individual variation among clinicians injects variability. Lack of consensus on clinical measurements of findings and diagnoses reduce the ability to collect valid data. For instance, the ambiguity of the definition of periodontitis and the methods to measure it (55) make it unclear how EDR data relate to periodontal disease entities, and to what extent these data are accurate and comparable across systems and providers. In addition, data accuracy and reliability are not routinely assessed in clinical care, making the measurement of error difficult.

Design and configuration of electronic record systems constrain data collection

As one of our previous studies (56) has found, EDRs often constrain the clinician’s ability to record specific data. EDRs accommodate some clinical data only poorly or not at all, forcing clinicians to abandon recording or find workarounds. Even if data fields exist, they may not capture data as required for a research study. Second, the process of research data capture is not integrated into EDRs (57). Last, electronic record systems are typically not interoperable, requiring significant effort to synthesize data about individual patients (58).

While reuse of electronic data for dental clinical research is still in its infancy, it is clear that healthcare as a whole is moving towards exploiting clinical data as a source of new knowledge (3). While clinical data suffer from multiple limitations compared to the research data collected in rigorously conducted studies, they are a potentially valuable resource that can complement and enrich the insights traditional research is generating. Part of our current difficulties is that we have never reused clinical data in dentistry broadly and for a variety of topics. Now that we are beginning to do so, we find that many assumptions about the validity, accuracy, reliability and utility of these data may be flawed. However, only through reuse and analysis of these data will we find ways to improve their usefulness for research.

LIMITATIONS

This review has two main limitations. First, our search strategy may not have identified all possible studies of interest due to limitations of database coverage and idiosyncrasies in article indexing. We attempted to compensate for this potential limitation by hand-searching key journals and using a snowball method. We believe that our detail-oriented approach identified as many studies as was possible. Second, some of the studies did not provide detailed descriptions of their data sources. In addition, authors used various terms to describe electronic data sources, such as computerized treatment database, patient database, or practice management systems. This circumstance sometimes impeded exact categorization of particular studies.

A LIST OF ABBREVIATIONS

EHR

Electronic health records

EDR

Electronic dental records

NDPBRN

National Dental Practice-based Research Network

NIDCR

National institute of Dental and Craniofacial Research

PBRN

Practice-based research network

EMR

Electronic medical records

ONJ

Osteonecrosis of the jaw

VA

Veterans Affairs

OSA

Obstructive sleep apnea

AO

Atypical odontalgia

CHD

Coronary heart disease

DMF

Decayed, missing and filled teeth

HMO

Health maintenance organization