David C. Grossman, MD, MPH

 To examine uptake patterns of the electronic health risk assessment (eHRA) and compare characteristics of early adopters among adults in 1 healthcare system with those of the potentially eligible population.

 Retrospective cohort study (September 2006-March 2009).

 We designed and implemented an integrated eHRA to improve individual health plan members’ self-management of their health risks and chronic illness. We included individuals aged 21 to 85 years who had been enrolled at Group Health Cooperative for 12 or more months before September 2006, when the eHRA was first introduced. Study participants had to be registered users of the health plan’s secure Web portal by March 2009 in order to complete the eHRA.

A total of 332,381 adults were potentially eligible; of these 39.3% were eligible to complete the eHRA. One or more eHRAs were completed by 22.4% of registered Web portal users over the study period with slow but consistent uptake. Completers were more likely to be women, to be middle-aged (41-65 years), and to have had a recent well-care visit and fewer comorbid conditions.

Significant additional outreach, engagement strategies, and incentives are likely required by health systems to increase adoption and ongoing use of an eHRA among target populations. Future research on eHRA uptake in primary care should focus on whether the use of these tools leads to patient action and provider engagement that improve health outcomes in moderate-risk and  high-risk individuals, as well as on modalities to reach broader audiences for higher completion rates.

The demographic and health risk profile of early electronic health risk assessment (eHRA) adopters was primarily characterized by age (41-65 years), sex (women), recent well-care visit, and fewer comorbid conditions.

		Significant additional outreach and engagement strategies and incentives are likely required to increase eHRA adoption and use among target populations and may decrease demographic differences between early adopters and later adopters.

		Future research on uptake of risk assessment tools in primary care should also address whether their use leads to increased uptake of activities that improve health outcomes in moderate-risk and high-risk individuals.

Unhealthy lifestyles have a substantial impact on the incidence of chronic conditions and impaired health outcomes for US residents.

Rising healthcare expenditures are associated with increasing chronic disease incidence and prevalence, as exemplified by the association of obesity and increased healthcare  costs.

 Some of these conditions or complications can be prevented or averted by reducing behaviors that increase health risk such as the use of tobacco, poor diet, physical inactivity, and risky alcohol use.

 Healthcare providers, and the systems in which they work, are well positioned to systematically identify high-risk individuals and provide assistance for behavior change leading to improved health and reduced healthcare costs associated with the downstream chronic illness.

Increasingly, innovative healthcare systems use population management strategies to identify at-risk individuals to provide behavioral interventions,self-management programs, and clinical interventions to reduce health risks. Helping people identify their risky  behaviors can contribute to better health status when linked with targeted wellness programs.

 One approach that is gaining  momentum has been for employers and healthcare organizations to administer health risk assessments (HRAs) to individuals, often electronically. The information collected is used to identify risk factors to provide tailored feedback and programs aimed at  changing unhealthy behaviors and promoting evidence-based screening, immunization, and preventive medications.

 Several systematic reviews on the effectiveness of HRAs concluded that when used alone, there was limited evidence on their effectiveness, with a broad range in evidence quality.

 However, the prevailing view is that HRAs, coupled with additional interventions including worksite health promotion activities and primary care interventions, have the potential to improve population health and to be cost-effective mechanism for comprehensively and systematically improving health outcomes.

New rules from the Centers for Medicare & Medicaid Services (CMS) call for the use of HRAs as a key part of the Medicare Annual Wellness Visit and require providers to use an HRA for prevention planning. Because of this rule, HRAs are likely to become much  more common in clinical practice.

Group Health Cooperative, an integrated health plan and delivery system based in Seattle, Washington, developed an interactive, online electronic HRA (eHRA) in September 2006.

 Through a secure member website (MyGroupHealth, www.ghc.org), the  comprehensive risk assessment tool collects self-reported information from adult members of all ages and integrates it with laboratory and biometric data from their electronic medical records (EMRs). The eHRA was developed as a new feature of  MyGroupHealth, the health system’s full-featured secure patient Web portal designed to provide patients with access to personalized health information and a mechanism for communication with their clinical teams.

Tailored feedback for behavior  change, preventive services, and chemoprevention, as well as targeted referrals to wellness programs, are fed directly back to patients and their clinical teams after eHRA completion.

This study was designed (1) to examine the adoption of the eHRA and compare the characteristics of early eHRA adopters with  those of the potentially eligible population and (2) to understand adoption trends in the setting of an integrated health system. Understanding the characteristics of early eHRA adopters is critical for downstream evaluation of the effectiveness of eHRAs on  outcomes and for refining strategies to boost outreach and engagement for providers and healthcare systems to reach nonresponders.

The retrospective cohort study was conducted at Group Health Cooperative between September 2006 and March 2009. Group  Health is an integrated healthcare system that currently serves approximately 675,000 patients in Washington State. Approximately  two-thirds of members receive primary and some specialty care in the 25 medical centers from a single large multispecialty physician group practice. The remaining “network” members receive their healthcare from contracted providers. All members have the opportunity, and are encouraged, to sign up for free access to the electronic patient portal,

 which is integrated with their EMR in the Group Health medical centers but not in the network.

 The Group Health Human Subjects Review Committee approved all study procedures.

Potentially eligible individuals were the 332,281 members aged 21 to 85 years who had been enrolled in one of Group Health’s medical centers for at least 12 months before September 2006 (baseline). To complete the eHRA, members had to be registered with the secure patient Web portal

; 39.3% (130,735) had registered access to the Web portal and were eligible to complete the  eHRA by March 2009. Completers were defined as individuals who were registered to use the patient Web portal and who  completed the eHRA anytime between September 2006 and March 2009; noncompleters were individuals who were registered to use the patient Web portal but who did not complete the eHRA by March 2009.

Group Health electronic enrollment and claims databases were used to collect demographic variables (age, sex), RxRisk  comorbidity score,20 well-care preventive visit in the year before baseline, duration of health plan enrollment before baseline, and is enrollment or death during the study period. Among individuals who completed at least 1 eHRA, we collected the first completion rate and the total number completed. We used self-reported eHRA data to provide more detailed health information on completers.

The eHRA was designed to be completed in 15 to 20 minutes through the secure Web portal. The extensively branched algorithm  resents questions on medical history and health-related behaviors, demographics, functional and  self-reported health status, social  and occupational history, medication use, and prior use of recommended preventive services (maximum number of questions was   63). After completion, members and their clinical teams receive personalized reports with tailored feedback on health status, risk  estimations for cardiovascular and other chronic diseases, screening and immunization recommendations, and lifestyle behavior  assessment and recommendations. Tailored educational material and Web links to specific resources (eg, tobacco cessation  rograms, lifestyle coaches) are also provided. A complex algorithm that accounts for the constellation of risk factors, demographics, medical history, and readiness to change is used to generate recommendations.

 which are  in large part based on the those of the US Preventive Services Task Force  recommendations, form the basis for many of the algorithms to provide recommendations about screening, immunization, chemoprevention, chronic disease management, and health behavior change.

Self-perceived health status, lifestyle risks, and chronic illness control (diabetes, hypertension, asthma, depression, heart failure) were assessed in the eHRA using standard approaches and measurement tools. Biometric data from the EMR (glycated hemoglobin and blood pressure) were prepopulated in the eHRA, but individuals had the ability to alter these imported numbers, so  we refer to all these data as self-reported. Self-perceived health status was collected with a single question and rated as excellent,  ery good, good, fair, or poor.

Lifestyle factors included body mass index (kg/m

), physical activity, tobacco use, and alcohol use. The International Physical  Activity Questionnaire was used to classify individuals as active, minimally active, or inactive.

 No lifestyle changes were  ecommended for nonsmokers or ex-smokers who had quit during or before the past year, whereas change was recommended for  urrent smokers. The Alcohol Use Disorders Identification Test

 was used to identify individuals with possible alcohol abuse or dependence and those with risky drinking behaviors.

The eHRA also evaluates control of diabetes, depression, and hypertension among persons reporting having these conditions. Diabetes questions were presented to every member who reported a personal history of diabetes (except for gestational or  secondary diabetes), and control was categorized using clinical laboratory data on glycated hemoglobin levels. Individuals who  eported seeing a counselor or taking antidepressant medications completed the first 8 questions of the Patient Health Questionnaire 9 tool.

As a strategy to promote uptake, the eHRA was promoted to primary care providers as a tool to prepare for adult prevention visits.  Rather than a typical paper questionnaire, patients were encouraged to complete the eHRA before a scheduled  well visit. During the study period, all Group Health clinics were provided financial incentives to promote eHRA use. The 3 clinics with the highest  esponse rates received a clinic-level cash reward for discretionary use. No financial incentives were offered by the health plan or the  medical group to patients or individual providers to complete an eHRA.

Individuals could have completed more than 1 eHRA in the study period; only the first completed questionnaire was included to  compare completers with noncompleters. We characterized individuals who completed more than 1 HRA during the study period.  data from the 2007 Group Health Wellness Inventory

 were used to compare the prevalences of health conditions reported by the completers. The Wellness Inventory included a stratified random telephone survey of 590 adult Group Health members and obtained  prevalence estimates on demographics, health conditions, and lifestyle risk factors. We used descriptive frequencies to compare completers with noncompleters on demographic and health services data with Pearson’s x

 test for categorical variables and means and standard deviations for continuous variables using SPSS version 10.1 (SPSS Inc, Chicago, Illinois). Significance was defined as 2-sided 

Of the 332,381 adults potentially eligible for inclusion in the study based on age and duration of enrollment before implementation of  he eHRA, 29,321 (8.8% of all potentially enrolled and eligible subjects and 22.4% of registered Web portal users) completed 1 or  ore eHRAs (

). The average enrollment duration before the start of the study was 10.3 years (standard deviation = 6.77 years). During the study period, 16% of members disenrolled or died (only 8.1% of completers). Patient enrollment in the Web portal  teadily increased between September 2006 (29.7% enrollment) and March 2009 (39.3% enrollment).

Comparison of Health Profile Completers and Noncompleters

The eHRA completers were more likely than noncompleters to be female (64.6% vs 56.9%), to be middle-aged (41- 65 years, 64.3%  vs 59.2%), and to have had a well-care visit (31.3% vs 27.7%). Based on comparisons from the Wellness Inventory, there  as no difference between completers and Group Health members as a whole in health status, body mass  index, physical activity,  or diabetes; in contrast, completers were less likely to be current smokers (8.1% vs 15.7%

The majority of respondents indicated their health was good (37.0%), very good (39.3%), or excellent (13.1%) (

).  Approximately one-third of respondents fell into each body mass index category. One in 5 (21.8%) reported they were physically inactive. Fewer than 10% (8.1%) were current smokers or had a moderate (4.8%) or high (0.6%) Alcohol Use Disorders Identification  est score. Diabetes was  reported by 7.6% of the respondents. Just over 15% reported being depressed; among  hese, 53.8% had mild, 24.5% moderate, and 21.0% severe depression. Hypertension was slightly more prevalent (17.8%), with  41.0% having poor control (>140/90 mm Hg or >130/80 mm Hg with diabetes).

Among individuals who completed any eHRA during the study period, 17% had completed it within 3 months of its implementation, 4% within 9 months, and 66% within 15 months (

). A minority (17.6%) of completers completed 2 or more eHRAs.

The eHRA uptake rate was slow but reasonably constant over implementation and resulted in just over 20% of individuals with  registered Web portal access completing the eHRA over a 31-month period (8.8% of all potentially enrolled and eligible subjects). Understanding whether individuals who complete eHRAs are representative of underlying populations is relevant for several reasons. First, if eHRAs are to be used to characterize the health status of enrolled populations, it is important to understand how individuals who complete these assessments differ from those who do not; without this knowledge, health systems could make a biased  assessment of the health status of their covered populations and could poorly target resources. Second, understanding selection  factors for completion will be critical for assessing whether use of these tools leads to improved health outcomes and population health. Finally, characterizing individuals who do not complete these tools provides an opportunity for reaching broader audiences for higher completion rates.

 we found that women, middle-aged individuals, and individuals with recent well-care visits and fewer  omorbid conditions were most likely to complete the assessment. Possible reasons for these findings are that younger people tend to be in good health and less concerned about their health status,

 while older people may feel that significant   improvements in health outcomes are not possible in the late stages of life.

 Or this finding may reflect patients’ interest in properly managing their risk factors.

During the study period, Group Health members had to carry out a 2-step process to complete their eHRA: first they had to sign up to use the secure Web patient portal and then  they had to sign onto the website to complete their questionnaire. Also at the time of  this study, the eHRA was only available in English, potentially limiting access for those with other primary languages. The  additional steps required to complete the eHRA could have further influenced the profile of completers, but appeared to have little  ssociation with who completed the eHRA among potentially eligible respondents. Despite these additional steps required to complete the eHRA, there were only a few notable differences in the prevalence of conditions and lifestyle risks between completers and Group Health members in general.

The eHRA represents an innovation in preventive care because it uses self-reported data on health risks and chronic condition management to provide recommendations that are shared with the patients and their providers and healthcare teams by integrating information from the Web portal with the EMR. Paper HRAs or eHRAs can be used to assist with clinical management by providers  and population management by medical groups and health plans; they also can be used by employers to improve population health.  Risk stratification of populations requires comprehensive diagnostic information, which includes integrating information from  diagnoses, laboratory values, pharmacy fills, and prior use patterns. For HRAs to improve population health, there needs to be broad uptake by patients to augment the medical record data with self-reported data and their use needs to be tied to patient and provider action that leads to improved outcomes.

Though eHRAs are not a new concept in clinical preventive care, their use has not taken root systematically in most healthcare systems. Health risk assessments have been used extensively by employers as part of worksite wellness programs to promote  health risk reductions among employees. The Community Preventive Services Task Force endorses the use of eHRAs as part of  these programs.

 In 2012, as part of new guidance regarding the requirements for annual wellness visits,

the CMS has  stipulated that a comprehensive HRA should be offered as a routine part of covered annual wellness visits for all Medicare beneficiaries. However, little is known about the characteristics of people who voluntarily complete these assessments and how they differ from the characteristics of people who do not. It is an important priority to evaluate the types of additional training and   resources that are needed by healthcare teams and systems to use HRAs to improve patient outcomes needs. Another high priority is to examine whether HRAs can provide actionable information for healthcare teams to improve health outcomes through patient action and provider engagement.

Financial incentives have been shown to improve uptake of risk assessment tools in worksite settings, but far less is known about  the role of incentives in delivery systems.

 We found the clinics with the highest response rates used direct outreach from the  physician’s office (telephone or e-mail) to patients, requesting completion of the eHRA as part of clinical care and population  management. Most of the individuals completed their first eHRA in the fourth quarter of 2006 and the third quarter of 2007, as the result of special promotions by Group Health Cooperative. The first peak (3-5 months after implementation) corresponded to the introduction of the Health Profile Questionnaire within Group Health staff. The second peak (13-15 months) corresponded to a clinic-level contest giving extra incentives for completing questionnaires. While these types of incentives may increase uptake, it is unknown whether increasing uptake alone will lead to improved receipt of preventive services and improved overall outcomes. 

Significant additional outreach, engagement strategies, and incentives are likely required by health systems to increase adoption  and ongoing use of an eHRA among target populations. Among patients in an integrated health system, the demographic and health risk profile of early eHRA adopters (completers) was primarily characterized by age and sex, a recent well-care visit, and fewer comorbid conditions. These types of risk assessment tools have the potential to address and integrate the interests of  patients and other stakeholders, including employers, clinical teams, and health plans, as long as they can provide actionable information for patients and healthcare teams with linkages to effective programs to mitigate health risks. Future research on the uptake of risk assessment tools in primary care should also address whether the use of these tools leads to increased uptake of activities that improve health outcomes in moderate-risk and high-risk individuals.

From Group Health Cooperative (DSMB, NKR, RJR, DCG), Seattle, WA.

Data collection was supported by the Group Health Research Institute’s Development Fund.

 The authors (DSMB, NKR, RJR, DCG) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Articles

Significant additional outreach and engagement strategies and incentives are likely required to increase adoption and ongoing use of health risk assessment tools among target populations.

Electronic Health Risk Assessment Adoption in an Integrated Healthcare System

 To determine if poor well-child care (WCC) visit adherence is associated with increased risk for Ambulatory Care—Sensitive Hospitalizations (ACSHs) among young children in an integrated healthcare delivery system.

This was a retrospective observational study.

We used claims and administrative data for children aged 2 months to 3.5 years enrolled at Group Health Cooperative from 1999 to 2006. Our main independent variable was timely WCC visits based on Group Health’s 2000 recommended schedule. We used Cox proportional hazard regression models to determine the association between WCC visit adherence and risk for a child’s first ACSH.

 Of the 20,065 children, 797 (4%) had an ACSH. Children with lower WCC visit adherence had increased hazard ratios (HRs) of 1.4-2.0 for ACSH (adherence 0-25%: HR 2.0, 95% confidence interval [CI]: 1.6-2.6, P <.001; adherence 26-50%: HR 1.4, 95% CI: 1.1-1.8, P <.05). Of the 2196 children with >1 chronic disease, 189 (9%) had an ACSH. Children with >1 chronic disease and with lower WCC visit adherence also had increased HRs for ACSH (adherence 0-25%: HR 3.2, 95% CI: 1.8-5.6, P <.001; adherence 26-50%: HR 1.9, 95% CI: 1.2-3.2, P <.05).

: For young children, poor WCC visit adherence was associated with increased risk for ACSH in this integrated healthcare delivery system.

Resources should continue to be directed at encouraging attendance at well-child care visits, especially among children with chronic disease.

		Prior literature has not evaluated this relationship in integrated healthcare delivery systems.

		This supports healthcare reform policies that provide preventive services free of costs.

		Well-child care visit adherence may be a useful marker to identify children that might benefit from a case management intervention

Children less than 5 years old have higher hospitalization rates than older children.

 Finding approaches to minimize morbidity for these children (eg, decreasing asthma exacerbations) may reduce hospitalizations. Regular attendance at well-child care (WCC) visits may prevent unnecessary hospitalizations through timely immunizations and/or chronic disease management. WCC visits likely have a higher impact for young children due to the greater frequency of recommended WCC visits in the first 3 years of life (13 visits) compared with older children (1 visit per year through age 18 years).

Prior studies evaluating the association between WCC visit adherence and hospitalizations have been mixed, as 3 studies found that higher WCC visit adherence was associated with decreased ambulatory care—sensitive hospitalizations (ACSHs) while 2 studies found no association.

 Only 3 of these studies accounted for the timeliness of WCC visits.

 No study has evaluated this relationship in integrated healthcare delivery systems. Unlike other healthcare systems, integrated healthcare delivery systems have greater control in standardizing and improving the care patients receive (ie, population-based care), which can decrease practice variation.

In addition, integrated healthcare delivery systems often place high importance on preventive care and therefore encourage and provide these services at no charge. Evaluating WCC visit adherence and risk of hospitalization in this setting provides a unique opportunity to understand this relationship where all patientsreceive similar WCC visit content.

The goal of this study was to understand whether parents’ timely adherence to WCC visits was associated with risk of ACSHs  among young children enrolled in an integrated healthcare delivery system. ACSHs represent hospitalizations which are potentially preventable with adequate access to and/or provision of outpatient care.

 We hypothesized that poor WCC visit adherence would be associated with increased risk for ACSHs in this integrated healthcare delivery system.

This was a retrospective cohort study conducted in collaboration with Group Health Cooperative, an integrated healthcare delivery system with ~640,000 members in Washington and Idaho. Claims from both the Group Health delivery system and contracted network physicians were included. This study was approved by both the Seattle Children’s Research Institute and Group Health Research Institute’s Institutional Review Boards.

We focused on children younger than 3.5 years because this age group is at higher risk for hospitalizations than older children,

because they have higher contact rates with healthcare providers, and because Group Health encourages 1 WCC visit between the ages of 2 and 3.5 years. Children were eligible for the study if they were enrolled by 2 months of age at GroupHealth between January 1, 1999, and December 31, 2006. Participants entered the study on either January 1, 1999 (ie, already a plan member), or on their first day of enrollment during the study period. When determining a child’s complete enrollment information, we considered gaps in enrollment of <45 days as continuous enrollment based on the Healthcare Effectiveness and Data Information Set definition.

 In addition, children needed to be enrolled on or prior to a recommended  WCC visit and through at least 1 of the subsequent recommended WCC visits in order to calculate adherence and minimize misclassification (

International Classification of Disease, Ninth Revision

) codes to identify children with chronic disease.13 Children were classified as having at least 1 chronic disease (ie, “

1 chronic disease”) if they had 1 or more claims matching an

 diagnosis included in the validated list. Children were classified as “healthy” otherwise. Since diagnosing asthma in children less than 4 years old can be challenging,

 were required for a child to be classified as having asthma. The top 5 chronic disease categories for our study were: asthma (38%), congenital heart disease (15%), congenital musculoskeletal anomalies (14%), failure to thrive (7%), and hereditary and acquired hemolytic anemias (6%).

A WCC visit was identified if an outpatient claim billed by a primary care provider (pediatrician, family physician, general  practitioner, osteopath, or nurse practitioner) contained the standard WCC visit 

 codes (eg, V20.2) in any diagnosis field. Adherence was based on Group Health’s 2000 WCC visit schedule, which was identical to the AAP’s schedule for the first 6 months and then differed from this schedule by requiring the following WCC visits through 3.5 years old: 10 months, 15 months, 2 years, 3.5 years (ie, the AAP’s 9-month, 12-month, 18-month, and 3-year visits are not required). WCC visits were included in the adherence calculation (numerator) if the WCC visit was “timely.” For example, a “4-month WCC visit” was allowed to occur on or after a child turned 4 months old and up to less than 6 months old (ie, the “4-to-6 month” WCC interval). To allow for an appropriate window of acceptability for when WCC visits might reasonably occur, we counted a “duplicate WCC visit” in 1 interval as the WCC visit for the following interval if the duplicate visit was <14 days (visits before 15 months of age) or 

30 days(visits after 15 months of age) prior to the start of a WCC visit interval where no visit had occurred. Otherwise only the first WCC visit was counted for WCC visit intervals with multiple WCC visits.

WCC visit adherence (range 0% to 100%) was a timevarying variable that was updated at the end of each WCC interval. For WCC intervals that overlapped completely with a child’s study period enrollment, WCC visit adherence was calculated by dividing a child’s total number of “timely” WCC visits by the total number of recommended WCC visits from start of enrollment through the end of each age-specific WCC interval (ie, a child enrolled for “X” number of WCC visit intervals would have “X” number of WCC visit adherence values for our regression analysis). We examined WCC visit adherence using a priori categories to facilitate interpretation and comparison with our prior study as follows: 0% to 25%, 26% to 50%, 51% to 74%, and 75% to 100% (referent).

Ambulatory Care—Sensitive Hospitalizations

As defined by the Agency for Healthcare Research and Quality, a hospitalization was classified as an ACSH if the primary diagnosis was one of those included in the predetermined list of ACSH diagnoses. The only exceptions to this were for the diagnoses of dehydration or iron deficiency anemia where the ACSH diagnosis could be either primary or secondary.

we excluded “adult” conditions (chronic obstructive pulmonary disease, congestive heart failure, hypertension, pelvic inflammatory disease, skin grafts with cellulitis, and angina). We also excluded “congenital syphilis” as these hospitalizations are unlikely to be affected by WCC visits due to the fact that these infections begin in utero or at birth. Additional hospitalization diagnoses were included based on their clinical importance to children less than 3½ years old(“pneumococcal meningitis” [

 320.2], and “septicemia due to H. influenza” [

 038.41]). Hospitalizations prior to the age of 2 months were excluded. Children could only have 1 ACSH for analysis since we were interested in whether WCC visit adherence was associated with the time to the child’s first ACSH.

Bivariate analyses consisted of Student’s 

 test for comparisons of continuous variables and the Pearson’s χ

 test for categorical variables. We used a time-varying Cox proportional hazards regression model to determine the association between WCC visit adherence and time from enrollment to the first ACSH. Children were censored when they had their first ACSH, reached the end of the study (December 31, 2006), disenrolled, or turned 3.5 years old, whichever came first. Potential confounders were included a priori in multivariate analyses based on the existing literature: patient age at start of enrollment,

 Age at start of enrollment was modeled as a continuous variable. The reference categories were male (vs female), no chronic disease, and never on Medicaid for our categorical variables of gender, chronic disease, and Medicaid enrollment, respectively. The proportional hazard  assumption was examined using STATA10. Statistical significance was determined at the 

We also conducted 2 sensitivity analyses. First, we used propensity scores in an attempt to control for self-selection bias (eg, children at greatest risk for ACSH may also be less likely to attend WCC visits).

 We predicted WCC visit adherence  (categorical) propensity score probabilities with multinomial logistic regression using age at start of enrollment, chronic disease status, and whether the child was ever enrolled in Medicaid. This sensitivity analysis (data not shown) did not change our results or conclusions. Second, although we could not adjust for an appropriate measure of continuity of care for this healthcare setting (ie, “clinic level”), we adjusted for continuity of care (COC) at the “provider level” based on the importance of this potential confounder from prior studies.

 We used Bice and Boxerman’s COC index (range 0 to 1) to measure how often children saw the same provider.

 We analyzed COC as a time-varying categorical variable that was updated at the end of each agespecific WCC interval. The magnitude of associations and confidence intervals (CIs) were similar to our primary findings (data not shown).

Of the 23,579 children that enrolled in the health plan prior to age 2 months, 20,065 (85%) children were eligible for analyses (Figure). Nearly a quarter of our population was on Medicaid and children with 

1 chronic disease accounted for approximately 11% of our study population (

). Based on when children exited the study, approximately threefourths of children fell into the highest WCC visit adherence category (75-100%, Table 1). WCC visit adherence was calculated based on at least 4 recommended WCC visits for 69% of children (median 5, interquartile range: 3-6).

During the study period, 797 (4%) children had at least 1 ACSH (Table 1). The median age at time of hospitalization was 13 months (interquartile range [IQR] 8-20 months) and 70% of all ACSHs were accounted for by the top 4 diagnoses (

). For healthy children, the top 5 ACSHs accounted for 65% of their ACSH (17% pneumonia, 15% dehydration, 14% asthma, 10% gastroenteritis, and 9% urinary tract infections). For children with 

1 chronic disease, the top 5 ACSHs accounted for 79% of their ACSHs (31% asthma, 15% pneumonia, 12% seizures, 11% gastroenteritis, and 10% dehydration). The median time from enrollment until first ACSH increased with increasing WCC visit adherence categories (

). The unadjusted rates of ACSH for children with one of our top 5 chronic disease categories were 9% for asthma, 11% for congenital heart disease, 8% for congenital musculoskeletal anomalies, 8% for failure to thrive, and 8% for hereditary and acquired hemolytic anemias. Lower WCC visit adherence was significantly associated with increased hazards for ACSH after controlling for presence of chronic disease, gender, enrollment in Medicaid, and age at start of enrollment (

). Compared with children with higher WCC visit adherence (75-100%), children who missed half or more of the recommended WCC visits were at 1.4 to 2 times the hazard foran ACSH (adherence 0-25%: hazard ratio (HR) 2, 95% CI:1.6-2.6, 

 <.001; adherence 26-50%: HR 1.4, 95% CI: 1.1-1.8, 

1 chronic disease who missed half or more of the recommended WCC visits were 1.9 to 3.2 times the hazard for an ACSH (adherence 0-25%: HR 3.2, 95% CI: 1.8-5.6, 

 <.001; adherence 26-50%: HR 1.9, 95% CI: 1.2-3.2, 

 <.05) compared with those in the highest WCC visit adherence category.

In this study evaluating the relationship between parents’ timely adherence to WCC visits and hospitalizations in an integrated healthcare delivery system, we found that young children who had lower WCC visit adherence (

50% adherence) had up to 2 times the hazard for ACSH compared with children with higher WCC visit adherence (>75% adherence). This relationship

was stronger for children with chronic disease and was surprisingly similar to our prior study in a traditional fee-for-service system.

This study adds to the limited and mixed evidence supporting WCC visit adherence and confirms our Hawaii study.

 In Hawaii, children typically are cared for by solo practitioners, which can increase the variability of care betweenhealthcare providers. In comparison, Group Health has greater control of variability through its population-based model of standardizing the care patients receive. Therefore, timely WCC visit adherence still appears to be an important component of reducing hospitalizations even in a system where patients are more likely to receive similar care for the same services.

Timely WCC visits, in theory, can reduce hospitalizations by providing opportunities to administer immunizations to prevent serious infections, proactively manage a child’s chronic disease, or enhance the primary care relationship. Previous studies have found that children with asthma who received better asthma management plans (eg, inhaled corticosteroids or asthma action plans) had lower rates of hospitalizations for asthma.

This may be particularly important since children with asthma have approximately 2.1 hospitalizations for every 100 children.

Unfortunately, a recent study found that only 46% of children with asthma received appropriate asthma care.

It is also possible that low WCC visit adherence may simply be a marker for other factors (eg, poor access, parental healthcare utilization beliefs, sociodemographic characteristics, having a trusting, well-established relationship with their child’s PCP) that are associated with increased risk for hospitalizations.

For example, parents of children who are less adherent to the WCC visit schedule may also be less adherent to other aspects of healthcare (eg, less likely to follow treatment regimens). If so, then improving WCC visit adherence may not necessarily reduce hospitalizations. However, WCC visit adherence could be used as a marker to identify children who may benefit from a case management intervention. WCC visit adherence is a relatively easy metric to monitor even in systems that only have access to claims data. Besides ensuring that children attend the recommendedWCC visits, health plan or primary care team case  managerscould discuss with parents whether they have the appropriate support for all of their child’s healthcare needs (eg, adequate transportation). From a population-based perspective, case managers can help health plans and providers ensure that their patients receive appropriate care.

Given the association found between increased timely WCC adherence and more favorable outcomes in this and prior studies,

 we should continue encouraging attendance at WCC visits. Similar to our study,

Hakim and Bye found that children up-to-date (vs not up-to-date) with WCC visit adherence had decreased hazards of avoidable hospitalizations.3 The only other study which accounted for timeliness of WCC visit adherence found no association with ACSH, which may be due to the relatively small number of ACSHs (N = 115).

 Although not statistically significant, these authors found that more children without an ACSH (63%) were up-to-date with WCC visits than those with an ACSH (53%, 

 One way of increasing WCC visit adherence might be to provide preventive services at no cost. Studies have found that adults with copays were less likely to receive recommended preventive care than those without copays.

During our study period, the majority of visits for WCC at Group Health did not require a copay. This might explain why WCC visit adherence, particularly for children with chronic disease, was higher in this integrated healthcare delivery system than in our prior study of children enrolled in a traditional fee-forservice

The limitations of this study are that we studied members enrolled in a single integrated healthcare delivery system and that children had high WCC visit adherence and were likely from higher-income families. Therefore our results may not be generalizable to other systems or populations. In this study, we could not adjust for some important potential confounding factors (eg, income, education, race/ ethnicity) due to limitations of available data. Future studies should attempt to directly  account for these potential confounders. We were also unable to control for continuity of care for this healthcare setting at the clinic level due to the limitations of the available data; during the study period, Group Health encouraged continuity at both the provider and clinic level. Therefore the magnitude of our findings may be biased away from the null. Due to the small number of children with chronic disease, these findings should be interpreted with caution. ACSHs are only “potentially preventable” and  therefore may not always be avoidable even with adequate access to and/or provision of outpatient care. In addition, some ACSHs may be related more to ambulatory specialty care (eg, seizures) than primary care. Our study design only suggests associations that cannot be inferred as causal. Although we attempted to control for self-selection bias, our available variables and the size of our study population may have limited our ability to adequately conduct propensity score analysis. Our study excluded 3514 (15%) children, although these children are unlikely representative of Group Health’s general population due to their very short duration of enrollment (median 1 month, IQR 1-2).

In this integrated healthcare system, timely WCC visit adherence was associated with reduced ACSHs. When compared with our prior study, our findings may help support healthcare reform policies that require preventive care free of costs. To better understand if this association could be causal, we recommend that future investigations rely on utilization data that are collected from electronic health records so that data from providers’ notes could be extracted. Our study findings could also be enhanced by qualitative data to further explore reasons why parents did not bring their child in for a WCC visit and survey data to collect important confounders (eg, income). In addition, future studies may want to evaluate emergency department visits as a mediator between low WCC visit adherence and hospitalizations or include diagnoses for acute injuries and poisonings, similar to a prior study,3 as these topics are often discussed at WCC visits. Until this association is clarified, poor WCC adherence can only be regarded as a potentially important risk factor for hospitalization among young children.

 Kaiser Permanente Center for Health Research (JOT), Honolulu, HI; Department of Pediatrics (JOT, RM-S, DCG), Department of Health Services (RM-S, DCG), University of Washington, Seattle, WA; Center for Child Health, Behavior, and Development (RM-S, CS), Seattle Children’s Research Institute, Seattle, WA; Group Health Research Institute (DCG), Seattle, WA; Department of Family Medicine and Community Health (C-WT), University of Hawaii, John A. Burns School of Medicine, Honolulu, HI; Pacific Health Research and Education Institute (C-WT), Honolulu, HI.

This study was supported by a Health Resources and Services Administration grant (T32HP10002-21), Group Health Cooperative, and Group Health Research Institute. Our sponsors did not have any role in the following: study design; the collection, analysis, and interpretation of data; the writing of the report; and the decision to submit the manuscript for publication.

 Dr Grossman reports that he is a shareholder in the Group Health Permanente Medical Group, which  contracts exclusively to provide medical care for Group Health members. Dr Tseng reports that she was supported by the Hawaii Medical Service Association Chair for Health Care Services Quality Research. The authors (JOT, RM-S, CS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (JOT, RM-S, DCG, C-WT); acquisition of data (DCG); analysis and interpretation of data (JOT, RM-S, DCG, CS, C-WT); drafting of the manuscript (JOT, RM-S); critical revision of the manuscript for important intellectual content (JOT, RM-S, DCG, C-WT); statistical analysis (JOT, CS); obtaining funding (JOT, DCG); administrative, technical, or logistic support (JOT, DCG); and supervision (RM-S, DCG).

 Jeffrey O. Tom, MD, MS, Kaiser Permanente Center for Health Research, Hawaii, 501 Alakawa St, Ste 201, Honolulu, HI96817. E-mail: jeffrey.o.tom@kp.org.1. HCUPnet: A tool for identifying, tracking, and analyzing national hospital statistics. http://hcupnet.ahrq.gov/. Accessed April 2, 2012.

2. AAP Recommendations for Preventive Pediatric Health Care (Periodicity Schedule). http://practice.aap.org/content.aspx?aid=1599.Accessed April 2, 2012.

3. Hakim RB, Bye BV. Effectiveness of compliance with pediatric preventive care guidelines among Medicaid beneficiaries. Pediatrics. 2001;108(1):90-97.

4. Pittard WB 3rd, Laditka JN, Laditka SB. Early and periodic screening, diagnosis, and treatment and infant health outcomes in Medicaidinsured infants in South Carolina. J Pediatr. 2007;151(4):414-418.

Timely well-child care visits may play a role in decreasing unnecessary hospitalizations in children in integrated healthcare delivery systems.

Well-Child Care Visits and Risk of Ambulatory Care-Sensitive Hospitalizations

To link the administrative data of a large dental carrier and an integrated health plan in Washington State to conduct an observational study of diabetes and periodontal disease.

 Evaluation of variable suitability, testing of linkage variables, and performing an 

We examined a variety of administrative data variables for their consistency over time and their information richness to use as matching variables. After choosing social security number, date of birth, first name, and last name, we tested their reliability as linking variables among a population with dual dental and medical insurance. Lastly, we performed four 

 − 1 deterministic linkage steps to obtain our study population.

 With a success match rate of more than 96% with the 4 test variables, we extracted the entire population who met the study criteria with the understanding that only a subset would successfully link. We linked 78,230 individuals (55.2% of the Group Health Cooperative population). Of these matches more than 50% occurred within a last name—first name–birth date deterministic match.

 Employer groups who provide dental-medical benefits for their employees send identical administrative data to dental and healthcare plans. The 

 − 1 deterministic linkage was accomplished by using a relatively straightforward approach because these data were fairly homogeneous and of high quality. Until medical care and dental care are integrated, it is possible to link these data to assess the impact of oral disease on overall health.

	Social security number, date of birth, first name, and last name were tested for reliability as linking variables among a population with dual dental and medical insurance.

		A simple n − 1 deterministic merge was used successfully to link the dental and medical records of individuals within a defined geographic region and a 5-year time interval, and with a permanent unique identifier in 1 of the datasets.

		Because insurance systems have generally evolved separately, record linkage is an important step to study the joint effects of medical and dental services on health and healthcare use.

	An emerging area of research interest is the intersection of medical and dental services, where the goal is to understand the potentially bidirectional relationships between dental and medical health, treatments, and outcomes. Evidence is accumulating on the association between dental conditions and diverse medical conditions including diabetes, cardiovascular disease, and chronic renal disease.

 As a result, expert groups and advisory bodies are recommending better integration of medical and dental care.

 However, medical and dental insurance and care systems have evolved separately with little integration, making it difficult to study the joint effects of medical and dental conditions. To conduct observational studies of the joint effects of medical and dental conditions, record linkage is required. Record linkage is a common tool for health researchers but has been infrequently used to link medical and dental care data.

	When no common identifier such as a medical history number is present to link data from unrelated organizations (as is the usual case for medical and dental records), the ability to successfully match  individuals from the databases of interest is strongly dependent on the amount, accuracy, and overlap of personal identifying information in the data files.

 Combining information in multiple, partially identifying data fields (eg, surname, first name, date of birth) can create linking algorithms with high sensitivity and specificity.

 Creating a successful linkage strategy depends on the number of linking variables, the information richness of those variables, and the accuracy with which they have been recorded. Having too few variables increases the likelihood of duplicate matches, and coding errors result in lost true matches. If substantial, such errors can result in misclassification bias in observational research studies.

	Deterministic matching and probabilistic matching are the 2 major methods of record linkage. At its simplest, deterministic linking requires an exact match on 1 or more identifiers between 2 databases. This linkage technique uses a reliable identifier such as social security number (SSN) and then performs verification using additional parameters. If records do not link, other relevant identifiers (eg, date of birth, name, sex) are used to iteratively link and verify the linkage. When reliable identifiers are available, deterministic approaches are usually preferred because they achieve acceptable results, are less cumbersome, and require less development time than probabilistic matching.

 The challenge is that across healthcare entities, identifiers (SSN, name, and address) may be coded incorrectly, different data standards or structures may be used, and different rules for data retention or updating may be applied, resulting in high nonmatch rates. For instance, in a Dutch study of perinatal care and pregnancy outcomes with 8 linking variables, complex probabilistic linkage techniques resulted in 80% more links than a deterministic linkage requiring exact agreement.

 To allow for minor discrepancies, some researchers allow for a less restrictive “

 − 1” deterministic linkage where 1 of the linkage variables is allowed to differ.

 Deterministic linkage rates also can be improved by simple methods to reduce uncertainty such as eliminating erroneous values, invalid dates, nicknames, prefixes, and suffixes.

	When linking data are poor or deterministic linkage achieves poor results, more complex probabilistic linkage techniques are needed. Probabilistic matching uses information from many variables and allows for disagreement. Agreement and disagreement weights are assigned to all possible record pairs depending on the degree of agreement and the number of different values possible. Records are declared linked if the total agreement weights achieve a predetermined threshold.

 Software packages are available to perform probabilistic matching when simpler deterministic matching strategies fail.

	The objectives of our study were to (1) quantify the prevalence of treated periodontal disease among diabetic patients age 40 to 74 years; (2) examine the association between periodontal disease control and diabetes control; and (3) quantify the difference in medical costs for diabetic patients who received periodontal services versus those who did not receive periodontal services. Our ability to conduct this study pivoted on our ability to link medical and dental records. We describe in this article the strategy used to link health and dental plan data from a large dental carrier and a large integrated health plan in Washington State. All study procedures were approved by the institutional review board at Group Health Cooperative (Group Health).

	Group Health is a nonprofit, integrated healthcare system that provides care to approximately 570,000 residents of Washington Sate and northern Idaho. Enrollment is through employer-sponsored programs, self-funded plans, individual and family plans, and Medicare and Medicaid plans. Washington Dental Service (WDS) is a dental benefits carrier in the state of Washington and provides comprehensive dental insurance to more than 2 million adults and children.

	For more than 20 years, Group Health has collected monthly administrative, financial, and clinical data formatted in SAS and relational databases. At Group Health, detailed demographic data are collected directly from enrollees or from their employers. We identified the possible linkage variables of sex, SSN, name, date of birth, residential address, and telephone number because these variables were common to both Group Health and WDS. At Group Health, enrollment information is updated by verifying the enrollee’s address and phone number at patient encounters, and changes in the enrollee’s name or SSN are submitted periodically by their employers. To maximize linkage rates, we extracted all variants of each linkage variable that were retained by Group Health over the 60-month study period. Washington Dental Service maintains 30 years of computerized dental history for all enrollees, which includes information on demographics, insurance coverage, providers, procedures, and claims. These data reside in an enterprisewide warehouse.

	At the time of enrollment, both Group Health and WDS assign a unique patient identifier to each enrollee. At Group Health, this identifier is permanently retained for each individual across all enrollment-disenrollment cycles, whereas at WDS a new unique identifier is assigned when an enrollee changes employer group. The SSN, while collected, is not used as the principal personal identifier in either plan and is missing for nearly 100% of dependents of the WDS subscriber.

	To explore the linkage utility of personal identifiers in both databases, we initially examined 10 potential linkage variables for missing or invalid values, and consistency over time. For the 200,864 Group Health enrollees born between January 1, 1932, and December 31, 1961 who were enrolled in both 2004 and 2005, we examined the availability and appropriateness for each field, and the percentage that changed from 2004 to 2005 (

). Last and first names were “cleaned” by removing spaces, suffixes, prefixes, and punctuation. None to very few missing or erroneous values were detected on identifiers that are verified at healthcare visits, including sex, last name, date of birth, first name, and address fields. Small numbers of missing or invalid values (<5%) were found for SSN and telephone contacts. The variables with the greatest number of unique values, and thus potentially the most discriminating, were SSN, telephone number, last name, first name, and date of birth. We found 49 variants for state, including military and other district codes (eg, Puerto Rico; Washington, DC) and 2 for sex, making them less useful as primary linkage variables. As expected, we found significant year-to-year variability in telephone number (17.7%), city (4.9%), and zip code (6.2%), making these variables less suitable for deterministic linkage. Enrollee first name, last name, date of birth, and SSN had less than 1% year-to-year variability, suggesting these variables were of high quality, and thus we chose them as our 4 linking variables. Social security number was included as it has evolved as a proxy universal identifier in insurance and healthcare systems and was expected to have high concordance between Group Health and WDS datasets.

	The second step was to test our confidence in the 4 linkage variables. After removing duplicates, we performed 4 deterministic linkages with 4 insured populations known to be dually insured by Group Health and WDS: Group Health’s Medicare Advantage plan, Group Health’s individual and family plan, and 2 large commercial accounts. We limited the study analysis to persons who were enrolled in 2004 or 2006 when the insurance products were active and who were born between January 1, 1932, and December 31, 1961. To improve matching precision, the Group Health and WDS datasets were standardized and names were cleaned. The linking results show 96% or higher exact matches among the 4 tests (

). Comfortable with the high degree of matching within the 4 fields we selected (SSN, last name, first name, and birth date), we developed the entire linked dataset for our study of diabetes mellitus and periodontitis. We included all individuals who met the study criteria from WDS and Group Health with the understanding that only a subset of the population would successfully link. First, we performed 2 exact deterministic linkages as a comparison group. An 

 − 1 deterministic linkage strategy was then performed on 4 sets of merges with 3 of the 4 merge variables: (1) SSN, last name, first name; (2) last name, first name, birth date; (3) SSN, first name, birth date; and (4) SSN, last name, birth date. To obtain our final linked population, we grouped the four 

 − 1 deterministic merges into 1 dataset and removed all duplicates using Group Health’s unique identifier.

	To protect confidentiality, all personal identifiers were stored on a computer file that only the Group Health analyst could access and were destroyed at the end of the study. Washington Dental Service sent SSNs separately from names and birth dates through a Web-based application (Secure File Transfer [SFT]). Access to this folder was restricted to the study analysts. Secure File Transfer uses the 128-bit Secure Sockets Layer encryption protocol that is the standard transmission protocol for secure information transfer over the Internet. User accounts within SFT are given specific and limited access to only the upload/download area associated with 1 study. The visibility or accessibility of files other than those of the specific study is prevented by the SFT design. Once the linkage was complete, personal identifiers were destroyed and replaced by study identifiers. To protect confidentiality and privacy, the study identifiers were used to combine medical and dental information on the same individuals. We used SAS software in all analyses (SAS Version 9.1 for Windows, SAS Institute, Cary, NC).

	Group Health has fewer enrollees than WDS, which is reflected in the total number of records pulled for our final merge: 413,954 for WDS versus 155,625 for Group Health. These totals are the entire population that met our study criteria.

	Among the 2 exact merges, SSN alone resulted in 41.4% linked records whereas the SSN, last name, first name, and birth date resulted in 36.3% linked records (

 − 1 deterministic linkages, the combination of last name, first name, and birth date had the highest linkage rate (50.6%). This result is explained by the nearly 32% (n = 132,032) of WDS enrollees whose SSN was missing because they were dependents of the primary subscribers; the SSN is not consistently collected for dependents. Fewer than 3% (n = 4035) of Group Health enrollees were missing the SSN. Although SSN is a unique identifier and is fairly accurate for linkage, if we relied on SSN alone we would have dropped 17,278 study subjects. The addition of SSN to our 

 − 1 deterministic model added an additional 2298 (3%) study subjects. Our final linked population (n = 78,230) is the result of grouping the four 

 − 1 deterministic merges into 1 dataset and removing all duplicates.

	The final linked population was 52% female, compared with 41% and 56% of the unlinked populations in WDS (18.3% had missing sex information) and Group Health, respectively. Similarly, the linked population had a mean (±SD) age of 52.1 years (7.5 years) compared with 49.9 years (6.7 years) and 52.9 years (8.0 years) for the WDS and Group Health unlinked populations, respectively.

	Although Group Health’s unique identifier allowed us easily to perform four 

 − 1 deterministic merges, it was not critical to the matching success. Fewer than 0.1% (n = 64) of the last name—first name–birth date merge were duplicates. Of these duplicates, 14 enrollees had 2 different last names, 27 had different first names, and 23 had different birth days. If we restricted Group Health data to the last enrolled date, there were only 3 last name–first name–birth date duplicates. However, we would have lost 226 matches (78,230 − 78,004).

	In this study we demonstrated a simple and successful approach to linking medical and dental records. The success of our linkage was based on a number of factors. First, we examined the reliability of many variables in the Group Health and WDS datasets and restricted matching to 4 identifiers with minimum errors or missing data. Second, we conducted test merges with these 4 selected variables among individuals known to have both WDS and Group Health coverage. Third, our time frame was short (2002-2006) and within a relatively small region (ie, Washington State). Thus, personal identifiers that are subject to change over time remained relatively constant. Fourth, Group Health maintains monthly administrative records for all enrollees. Thus, Group Health had records of all name variations, and all their variants were used to merge with the WDS dataset, increasing the likelihood of name matching. The Group Health unique identifier that is permanently assigned to each member was used to ensure no duplicate matches.

	Many studies that link disparate datasets have found that an individual’s first name varies considerably. Differences may present as a nickname, abbreviation, different spelling, or initial. Hence, first name often is an unreliable variable for linked datasets. Our insurance data were fairly homogeneous and of high quality and completeness. We did not experience much variance in the first name between WDS and Group Health datasets. We believe employer groups send identical administrative data to dental and healthcare insurers. For example, some first names were structured with an initial followed by a first name (eg, M Kay) in both datasets. Matches were high even in the absence of SSN. There are numerous resources that describe methods to handle problematic data.

	Medical care and dental care in the United States have evolved as separate entities, each with its own governing bodies, professional terminology, research interest, and insurance realms. As a consequence, integration in patient care is lacking, especially with respect to preventive and chronic care. Because a growing body of research associates chronic health conditions to chronic dental conditions, these heterogeneous databases need to be successfully linked to enable further research. Until medical and dental healthcare systems are integrated into a single model with 1 health data system, we will need to rely on successful computerized record linkage methods to further our understanding of the health implications of dental and medical care. In this article we show that a relatively simple deterministic approach, within a narrow time frame and with demographic variables commonly collected by dental and medical insurance carriers, results in high-quality record linkage. Further research is required to determine whether similar success rates can be obtained with enrollment data from other dental and medical care plans.

From Group Health Research Institute (MKT, RJR, KMN, LS, DCG), Seattle, WA; and Washington Dental Service (MC, REI), Seattle, WA.

This study was funded by the Washington Dental Service and Group Health Cooperative.

Ms Theis and Drs Reid, Newton, Spangler, and Grossman report receiving research grants from Washington Dental Service. Dr Inge and Ms Chaudari are employees of Washington Dental Service, which provided funding for this study.

Concept and design (MKT, RJR, MC, KMN, LS, DCG, REI); acquisition of data (MKT, MC, KMN); analysis and interpretation of data (MKT, RJR, MC, KMN, DCG, REI); drafting of the manuscript (MKT); critical revision of the manuscript for important intellectual content (MKT, RJR, MC, KNM, LS, DCG, REI); statistical analysis (MKT); obtaining funding (KMN, LS, DCG); and supervision (REI).

Robert J. Reid, MD, PhD, Group Health Center for Health Studies, 1730 Minor Ave, Ste 1600, Seattle, WA 98101. E-mail: 

1. Saremi A, Nelson RG, Tulloch-Reid M, et al. 

Periodontal disease and mortality in type 2 diabetes. 

An n - 1 deterministic linkage strategy was used successfully to merge dental and medical data from a healthcare plan and a dental insurance carrier.

David C. Grossman, MD, MPH

Articles

Electronic Health Risk Assessment Adoption in an Integrated Healthcare System

Well-Child Care Visits and Risk of Ambulatory Care-Sensitive Hospitalizations

Case Study of Linking Dental and Medical Healthcare Records