To evaluate the factors associated with nonadherence in a nationally representative sample of patients receiving lipid-lowering therapy (LLT).

 Retrospective database analysis of treatment-naïve (1 year without LLT claim) hyperlipidemia patients evidenced by a new pharmacy claim for lipid-lowering therapy.

 Pharmacy and medical claims data were analyzed for currently enrolled members receiving a new LLT from 2007 to 2008. Adherence was defi ned as percentage of days covered (PDC) and values <80% were used to categorize nonadherent patients. Independent variables included patient demographics, pharmacy utilization, and medical conditions. Stepwise logistic regression was used to predict the odds of nonadherence. Laboratory data variables were incorporated in an exploratory sub-analysis to test the robustness of the original model.

Adherence with LLT was estimated in 88,635 patients. Sixty-fi ve percent of patients were nonadherent (mean PDC = 0.33). Compared with statins, patients treated with bile acid sequestrants were 6.75 times as likely to be nonadherent (

 <.05) predictors of nonadherence included age 45 to 55 years (ref: age >75 y) (odds ratio [OR]: 1.11); prior diabetes diagnosis (OR: 1.15); and unique pharmacies used (OR = 1.10). Signifi cant factors reducing nonadherence include male gender (OR: 0.77); previous heart attack (OR: 0.82); prior adherent behavior (OR: 0.89); and unique physicians seen (OR: 0.97). Compared with no copayment, patients with $5 to $30 copayments had a signifi cant reduction in the likelihood of nonadherence.

 Medication adherence remains poor in patients receiving LLT. Treatment outcomes and healthcare resource use may be improved by prioritizing adherence programs in at-risk patient populations.

Adherence interventions may be best prioritized in populations at high risk for medication nonadherence. Compared with associated reference groups, our study suggests:

		Women, middle-aged patients, those treated with bile acid sequestrants, patients with diabetic and neurologic comorbid conditions, and those using multiple pharmacies had worse medication adherence.

		 Medication adherence was improved in men; patients with documented heart attacks; those with prior medication adherence; those with more routine physician use; and patients with copayments ranging from $5 to $30.

		 Laboratory data suggested patients furthest from serum lipid goals have the highest likelihood of medication nonadherence.

Over 800,000 Americans die from cardiovascular disease (CVD) every year. Approximately 75% of fatal cardiac events are due to heart attack (acute myocardial infarction [AMI]) and stroke (cerebrovascular accident [CVA]).

 Those who survive these events have a reduction in life expectancy of up to 15 years

 The direct and indirect societal costs attributable to CVD exceed $475 billion annually.

A substantial body of clinical research supports lipid-lowering therapies (LLTs) as the primary therapeutic modality for reducing the risk of cardiovascular outcomes. Statins are the primary treatment modality and may confer lipid reductions of 10% to 60%.

 In the primary prevention of CVD, lipid reductions with low to medium potency statins have been shown to reduce the incidence of fatal and nonfatal cardiovascular outcomes signifi cantly (by 30% to 40%).

High potency statins may deliver a larger reduction in the risk of AMI and CVA (36% to 48%).

 Depending on patient need, statins, as well as fibric acid derivatives, bile acid sequestrants, niacin preparations, cholesterol absorption inhibitors, and free fatty acid products, can be individualized to help meet clinical, economic, and humanistic goals.

There are significant clinical data demonstrating the effi cacy of LLT in CVD; however, translating the benefi ts from clinical trials into realworld outcomes is hampered, in part, by low medication adherence. Clinical trials demonstrate that adequate medication adherence over 5 years is necessary to reduce the negative clinical outcomes associated with CVD.

However, in clinical trials for LLT, full medication adherence is achieved through treatment oversight using clinical staff or by incorporating inclusion criteria that preselect patients on the basis of demonstrated medication adherence.

Real-world adherence with LLT does not typically achieve the levels observed in clinical trials. In a study of 34,501 Medicaid patients, medication adherence with LLT dropped from 45% to 36% 3 months after LLT initiation and 79% to 56% 6 months after initiation.

 Retrospective observational studies have found that women,

 nonwhite race, usage of a Medicaid plan (odds ratio [OR]: 1.60), and age >75 years were associated with medication nonadherence.

 Clinically, patients with depression and dementia,

a hospitalization in the year prior to the start of LLT,

 were associated with nonadherence. Treatment attributes including higher daily doses of LLT

 and medication switching reduced medication adherence.

 Across studies, signifi cant heterogeneity exists in the patient populations evaluated, model variables included, definitions of medication adherence, and reporting.

The purpose of this study was to study factors associated with nonadherence in a nationally representative sample of patients receiving LLT for dyslipidemia using several new factors not included in previous studies. This study described characteristics of adherent and nonadherent patients; identifi ed factors with signifi cant impact on nonadherence; and suggested strategies to operationalize these results.

Data for this study were derived from paid pharmacy and medical claims and serum lipid laboratory data from a commercial health plan with members in 14 states from 4 regions of the United States. Data covered January 1, 2006, to April 30, 2009. Patient inclusion criteria included: receipt of LLT (“index drug”) between May 1, 2007, and April 30, 2008; 1-year continuous eligibility before and after the date of index drug receipt (“index date”); and a 1-year period without an LLT claim prior to receipt of the index LLT. Medicare and Medicaid enrollees were excluded.

Pharmacy claims included demographic and pharmacy utilization information such as medications, costs, quantities, and days of supply. Medical claims captured 

International Classification of Diseases, Ninth Revision, Clinical Modification

) codes and their respective diagnosis dates. Serum lipid values were retrieved for a sub-sample of patients for whom laboratory data were available. Data were de-identified and the study was approved by the Institutional Review Board at The University of Southern California.

The unit of analysis was the patient’s first observed episode of LLT. Each patient’s pharmacy, medical claims, and laboratory data (where available) were summarized into 1 patient-level claims data set with multiple summary variables based on the index date of first LLT. The index drug was the key variable from which adherence estimates were derived.

Percentage of days covered (PDC) was used to estimate medication adherence. PDC was defi ned as the days' supply of LLT dispensed in the fi rst year following the index date divided by 365 days, multiplied by 100. The days’ supply of LLT was calculated as the total amount of medication provided divided by the frequency of administration (60 tablets, dosed twice daily equals a 30-day supply). PDC offers the advantage of measuring both medication adherence and persistence,

 both of which are important aspects of disease control in dyslipidemia. From a health plan perspective, the fixed observation period (1 year) used in PDC calculations aligns with a typical member enrollment period and payer decision-making timelines. PDC <80% was used as the cutoff for nonadherence with LLT and was the primary outcome variable.

Potential predictors of nonadherence included patient demographics; dichotomous variables based on selected 

codes; and variables capturing pharmacy and healthcare provider utilization in the year before the receipt of the index drug. Characteristics of the index drug such as copayment level and therapeutic class were also used as predictive covariates. Prior adherent behavior was defi ned as the ratio of total number of prescriptions filled and refi lled to the total number of unique medications used. A sub-analysis of patients with laboratory data was also performed. Evaluated here were the predictive value of baseline low-density lipoprotein (LDL) values; unique laboratory visits; and elapsed time between the most recent LDL laboratory and index date on nonadherence. This analysis was also meant to act as a sensitivity analysis to test the robustness of the full primary model.

The population was divided into adherent and nonadherent members, and descriptive statistics were compared using student t tests and 

 tests. PDC <80% was regressed on covariates derived from demographic, pharmacy, medical, and laboratory data using stepwise logistic regression. Multilevel categorical variables such as age categories were forced into the model if any 1 age category was found to be significant. Signifi cant predictors of nonadherence were identified at the 

 <.05 level. All statistical analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina).

. In total, 88,635 patients were identifi ed and 65% (n = 58,074) were nonadherent with LLT over the fi rst posttreatment year. Mean PDC was significantly (

 <.05) lower in nonadherent versus adherent patients (PDC = 33% vs 100%). Adherent patients were significantly older (53.7 vs 51.2 years) and male (55.6% vs 51.1%) compared with nonadherent patients. The number of comorbid conditions recorded in nonadherent patients was significantly greater than in adherent patients. Statins were the most commonly initiated treatment, and utilization was signifi cantly higher in adherent patients. Where available, baseline LDL values were significantly higher for nonadherent patients (144.5 vs 140.6 mg/dL).

The results of the stepwise regression of the predictors of nonadherence with LLT are displayed in 

. This procedure resulted in 34 predictors of nonadherence with differential levels of signifi cance.

There were 36 drug products used to initiate treatment for dyslipidemia which were further grouped into therapeutic class variables. Patients using bile acid sequestrants were 6.7 times as likely to be nonadherent as patients treated with statins (95% confi dence interval [CI]: 5.8-7.8). Patients using niacin products, free fatty acid products, cholesterol absorption inhibitors, and combination products were all significantly more likely to be nonadherent (

.002). The impact of treatment with fi brates on nonadherence was nonsignifi cant.

Compared with patients having a $0 copayment, patients with copayments of $5 to $10, $10 to $20, and $20 to $30 (taken together, $5 to $30) had a significantly decreased likelihood of nonadherence of 12%, 12%, and 6%, respectively (all 

 <.01). Copayment levels of $0 to $5 and greater than $30 also decreased nonadherence, but these effects were nonsignifi cant.

 A signifi cant gender effect was found: male patients were 23% less likely to be nonadherent versus females (

 <.001). Increasing age was found to decrease nonadherence. Patients aged 25 to 35 years were 26% less likely to be nonadherent versus those under 25 years of age. Patients over 75 years of age were 67% less likely to be nonadherent than the youngest patients. Patients located in the Central and East regions had roughly a 6% and 16% increased likelihood of nonadherence versus those in the West region, respectively (

Articles

MANAGERIAL

Medication adherence in hyperlipidemia remains poor on a nationally representative level. Predictors of medication adherence from a nationally representative commercial health plan are reported.

Factors of Hyperlipidemia Medication Adherence in a Nationwide Health Plan

 To predict adoption (80% rate of use) of an electronic health record (EHR) by admitting physicians using a heuristic model.

 Administrative data collected for 326 physicians who admitted at least 10 patients to 3 hospitals during the 6 months following EHR activation represented more than 80% of the total admissions. Functions evaluated included computerized physician order entry (CPOE), electronic history and physical (EH&P), and electronic discharge summary (EDS). Independent variables included hospital size, physician alignment, physician group size, use of an offi ce EHR, age, sex, specialty, volume, hospital based, inpatient to outpatient ratio, and loyalty.

<.01) for physicians who were employed, male, and had a high inpatient ratio, a lower patient volume, and a community hospital setting. EH&P and EDS adoption were more likely for physicians with financial alignment and a large academic hospital setting.

Personal factors (loyalty, age, sex) were generally not predictive. Organizational factors (hospital setting, financial alignment) were most predictive of adoption. Study results may help administrators improve EHR installations.

We used a heuristic model to predict adoption of a Healthcare Information and Management Systems Society stage 7 electronic health record (EHR) by 326 admitting physicians. Adoption of computerized physician order entry (CPOE), electronic history and physical, and electronic discharge summary was evaluated.

		 Physician-hospital financial alignment was the strongest predictor for all 3 functions, although hospital environment may affect each function differently.

		 Demographic factors, volume, inpatient ratio, group size, and loyalty were not consistently predictive of EHR adoption.

		 The benchmark of 80% use rate of CPOE is achievable in this environment.

Hospital electronic health records (EHRs) are thought to significantly improve healthcare processes, yielding safer, more cost-effective care.

 Electronic health records are complex, integrated systems and cost millions of dollars to develop and implement. Physicians are key users of the systems and hold signifi cant power to reduce hospital costs and improve quality,

 though physician nonuse of EHRs is a serious problem.

 We developed a pragmatic, predictive model for the adoption of a Healthcare Information and Management Systems Society stage 7 EHR using predictors at the hospital, physician group, and physician levels.

Although the cost of technological innovation has been blamed for more than half of healthcare cost increases,

 some innovations reduce the cost and improve the quality and safety of healthcare. Electronic health record systems are considered essential to future improvements in care.

In systematic reviews of the literature, health information technology was shown to improve quality by increasing guideline adherence, decreasing medication errors, and enhancing disease surveillance.

 Research also showed improved effi ciency through decreased utilization of healthcare resources such as redundant diagnostic tests. The integrated EHR—seamlessly connecting primary, secondary, and tertiary providers of care&mdash;is regarded as the ultimate goal, but has only been attempted by a few large health systems in the United States.

 Current research regarding the use of EHRs, however, “is scattered across many different clinical contexts and involves many different types of EHR.”

There are numerous functions included in an integrated EHR intended to improve the quality, safety, or effi ciency of care. Three functions recognized as important are computerized physician order entry (CPOE), electronic history and physical (EH&P) documentation, and electronic discharge summary (EDS) documentation.

CPOE, augmented by computerized clinical decision support, is one of the processes expected to immediately improve clinical outcomes.

 The percentage of orders entered directly by the physician via the computer is advocated by the Agency for Healthcare Research and Quality as an important measure of EHR use.

The benchmark CPOE rate of 80% has been advocated by several experts.

 The percentage of orders entered directly, however, varies widely from hospital to hospital and physician to physician.

There is a dearth of research about the adoption of the EHR documentation functions. Studies suggest structured data entry into the EH&P and EDS can prompt completeness, provide greater accuracy and readability, and improve communication. 

 Additionally, electronic capture of these data assists in coding and extraction of data for future health services research.

While it is generally assumed that EHR adoption will yield positive outcomes, extant research on factors infl uencing physician adoption of EHRs implemented by hospitals is biased by several factors, including low survey response rates and methods too cumbersome to be practical for use by healthcare administrators. Our research uses a pragmatic, “common sense” multilevel view of adoption of innovation, examining level of EHS adoption at a model healthcare system as a function of variables at 3 levels: hospital, physician group, and physician.

This study used retrospective administrative data to evaluate the rate of adoption of a Healthcare Information and Management Systems Society stage 7 EHR by physicians at 3 hospitals in an integrated healthcare system in Virginia. The EHR was interoperable between hospital, ancillary, and physician office locations. The cost of the EHR was about $30 million per hospital. The hospitals selected for evaluation activated the EHR within 9 months of one another. The software, hardware, and implementation and training methods were consistent and stable. The hospitals include 2 community hospitals and 1 large academic hospital in order to provide a better cross-section of a typical hospital system. The sample excluded low-volume admitters (fewer than 10 patients per 6 months), physicians who admitted patients during fewer than 3 of the 6 months evaluated, and physicians with prior exposure to the EHR.

What variables predict physician adoption of an EHR, with adoption defi ned as the initial achievement of 80% use within 6 months after EHR activation? Dependent variables included rates of CPOE, EH&P, and EDS. Independent variables were based on a heuristic combination of constructs from the diffusion of innovations theory and the resource dependence theory. Adoption was hypothesized to be positively associated with you ng age, higher volume, high physician-hospital alignment, high levels of loyalty, hospitalbased practice, medicine specialties, being from larger groups, having offi ce-based EHRs, admission of a higher proportion of inpatients versus outpatients, and admitting at larger teaching hospitals. Physician hospital alignment was divided into 4 categories: employed by the health system, aligned by a contract, independent, and competing (contracted or employed by a competing health system). No a priori hypotheses were specifi ed for differences between CPOE, EH&P, and EDS.

Because of the clustering of physicians within hospitals, models were estimated with generalized estimating equations with SAS/STAT software version 9.2 using PROC GENMOD. 

The strengths of the individual variables with respect to their individual odds ratios (ORs), their alpha signifi cance, and their QICu statistics

 were considered in the determination of the best models of rate of adoption. The common correlation for all of the generalized estimating equation models was 

 = -0.0057, suggesting minimal clustering effect. Power was calculated for the models using R2 software.

 The calculations suggest adequate power (>.80) for this sample size and number of variables.

The physicians included in the sample accounted for more than 80% (n = 84,326) of admissions to the 3 hospitals during the study period.

Separate models for CPOE adoption, EH&P adoption, and EDS adoption using all of the independent variables were fitted and subsequently reduced. Results for the bestfitting models are presented in the 

. For the CPOE full model the QICu was 264.98, and for the reduced model the QICu was 262.35. While there was not much change in QICu, the reduced model was slightly better fi tting. The likelihood of CPOE adoption was higher for physicians at the small community hospital (OR = 8.27) and the midsized community hospital (OR = 2.9) compared with the large academic hospital; higher for employed physicians (OR = 24.9) and contracted physicians (OR = 1.2) compared with competing physicians; higher for men compared with women (OR = 3.3); higher for surgeons compared with medical physicians (OR = 1.83); lower for physicians in the top quartile of discharges (more than 132) compared with the fi rst quartile of discharges (fewer than 36) (OR = 0.34); and lower for physicians in the second quartile of inpatient ratio (less than 46%) compared with the fourth quartile of inpatient ratio (more than 98%) (OR = 0.26). Age, group size, use of an office EHR, being hospital based, and loyalty were not significant predictors of adoption of CPOE by 6 months once adjusted for other variables

For the EH&P full model the QICu was 381.50, and for the reduced model the QICu was 377.53. While there was not much change in QICu, the reduced model was slightly better fitting. The likelihood of EH&P adoption was higher for the large academic hospital compared with the midsized community hospital (OR = 1.52) or small community hospital (OR = 2.94); higher for contracted physicians (OR = 20.5) and employed physicians (OR = 6.99) compared with= competing physicians; higher for women compared with men (OR = 1.7); and lower for physicians with discharges above the median (more than 70 discharges) compared with physicians with discharges in the fi rst quartile (ORs = 0.44 and 0.29).

For the EDS full model the QICu was 337.75, and for the reduced model the QICu was 328.47. Here, the QICu had a slightly larger change, indicating that the reduced model was better fitting. The likelihood of EDS adoption was higher for the large academic hospital compared with the mid-sized community hospital (OR = 2.9) and the small community hospital (OR = 1.3); higher for independent physicians (OR = 1.8), contracted physicians (OR = 12.7), and employed physicians (OR = 5.27) compared with competing physicians; higher for surgical specialists compared with medical specialists (OR = 2.3); and lower for hospital-based physicians compared with those who were not hospital based (OR = 0.45).

The combination of the diffusion of innovations theory and resource dependence theory provides a pragmatic model for the prediction of EHR adoption. Interestingly, organizational-level variables and level of physician-hospital alignment were more predictive than individual physician—level variables. The hospital type was signifi cantly associated with adoption of all 3 EHR functions. The direction of that association varied and may suggest that leadership, change management methods, social norms at each facility, or other environmental factors affect adoption differently for different functions in the EHR software.

Adjusted for other variables including hospital, the type of financial alignment was clearly the strongest predictor of adoption of all of the measures of EHR adoption. Employment and contractual alignment both increased the likelihood of adoption of all 3 EHR functions. Financial incentives or goals to meet the predetermined threshold of 80% use were known to be present in some of the contracts and employment agreements. Moreover, the lack of association between loyalty and adoption suggests financial relationships are more important than nonfinancial relationships. Additionally, admitting physicians employed by competing health systems may consider it an inconvenience to care for patients at the competing institutions, only admitting patients at the health system studied because of exclusive contracts with payers or because of patient request.

Another interesting finding is the difference in the speed and level of adoption of CPOE versus EH&P and EDS as shown in the 

. This suggests there is a difference in the way the physicians approach the “action” function of CPOE (telling people what to do) versus the “retrospective documentation” functions (telling people what was done) of EH&P and EDS. Another likely explanation may be the level of difficulty physicians experienced in using noncomputerized methods. Forms for written orders were removed from most areas of the hospital and verbal orders were strongly discouraged by the hospitals’ leadership, while the use of verbal dictation for the documentation of history, physical, and discharge summary information remained readily available. The effect of monitoring metrics and administrators’ attention to meeting goals may also explain some of the differences in adoption of CPOE compared with EH&P and EDS. Administrators emphasized CPOE compliance, evaluating use rates for each physician on a weekly basis. Use of the electronic documentation functions was encouraged but did not receive the same level of monitoring and feedback as CPOE.

The adoption of CPOE at a level of 80% was suggested by experts as the hurdle for the receipt of incentive payments from the government. The fi nal Health Information Technology for Economic and Clinical Health (HITECH) Act reduced the level to 50%. This study shows the 80% rate is achievable and provides information about the association of physician alignment and other variables. No incentives are in place for the use of the documentation functions of the EHR, and physician adoption of those functions lags behind that of CPOE. Since alignment was signifi cantly associated with adoption, it may be valuable for the government to continue plans to integrate the payment for services for physician and hospital services, thereby encouraging physician-hospital alignment and consequently EHR adoption.

Hospital administrators may use the results of this study to perform an assessment of the likelihood of EHR adoption by the medical staff at individual hospitals. Understanding the variables that are associated with success or failure will enable them to devise strategies to improve the odds of high use and rapid adoption. The effect of the hospital environment was significant but much less powerful than the effects of alignment when adjusted for type of hospital. Understanding the association of financial alignment may encourage administrators to further their efforts to hire or contract with physicians. Actions may also include increasing attention to physicians predicted to be slow adopters to increase the odds that they buy into the system. Hospital administrators traditionally have relied on good relationships and physician loyalty as a predictor of future behavior. Loyalty, however, was not a signifi cant predictor of adoption or the rate of adoption. Administrators should be aware that strong informal relationships do not predict adoption by admitting physicians, although formal financial alignment is a strong predictor.

Hospital and physician-hospital alignment, but not loyalty, are predictors of integrated electronic health record adoption by admitting physicians in an integrated system.

Predictors of Physician Use of Inpatient Electronic Health Records

Geographical barriers to subspecialty care may prevent optimal care of patients living in rural areas. We assess the impact of human immunodeficiency virus (HIV) and hepatitis C telemedicine consultation on patient-oriented outcomes in a rural Veterans Affairs population.

 This was a pre- and post-intervention study comparing telemedicine with in-person subspecialty clinic visits for HIV and hepatitis C. Eligible patients resided in 2 rural catchment areas. The primary binary outcome was clinic completion. We estimated a logistic regression model with patient-level fi xed effects. This approach controls for the clustering of visits by patient, uses each patient’s in-person clinic experience as an internal control group, and eliminates confounding by person-level factors. We also surveyed patients to assess satisfaction and patient-perceived reductions in health visit—related time.

 There were 43 patients who accounted for 94 telemedicine visits and 128 in-person visits. Clinic completion rates were higher for telemedicine (76%) than for in-person visits (61%). In regression analyses, telemedicine was strongly predictive of clinic completion (OR 2.2; 95% confi dence interval [CI]: 1.0-4.7). The adjusted effect of telemedicine on clinic completion rate was 13% (95% CI: 12-13). Of the 30 patients (70%) who completed the survey, more than 95% rated telemedicine at the highest level of satisfaction and preferred telemedicine to in-person clinic visits. Patients reported a signifi cant reduction in health visit—related time (median 340 minutes, interquartile range 250-440), mostly due to decreased travel time.

 HIV and hepatitis C telemedicine clinics are associated with improved access, high patient satisfaction, and reduction in health visit—related time.

We evaluated human immunodefi ciency virus (HIV) and hepatitis C telemedicine services at 2 rural clinics.

 HIV and hepatitis C are high-risk, prevalent, and chronic illnesses where subspecialty care can often improve quality of life and survival. Distance may be an important barrier to access.

 Visit completion rate, a proxy for access, was 13% greater for telemedicine compared with in-person clinics.

 >95% of respondents rated telemedicine at the highest level of satisfaction and preferred telemedicine to in-person visits.

 Telemedicine reduced health visit time by 340 minutes.

 This intervention was implemented within an existing telemedicine infrastructure without need for additional staff or equipment.

Electronic interactive healthcare consultations, also known as telemedicine, may reduce geographic barriers to timely subspecialty care. Human immunodefi ciency virus (HIV) and hepatitis C are ideal test conditions for telemedicine, because they are high-risk, prevalent, and chronic illnesses where subspecialty care can often improve quality of life and survival. The demand for such services will likely increase as more patients become eligible for treatment with new antiviral regimens.

 Unfortunately, limited access to subspecialty care is a major barrier to optimal care of patients infected with HIV and/or hepatitis C,

 particularly for those who live in rural settings.

The Veterans Affairs (VA) Healthcare System has taken a leadership role by investing heavily in telemedicine

 to better serve the 20% of veterans who reside in rural areas.

 Expanding telemedicine access to HIV/hepatitis C subspecialty care has been identified as a top priority by the VA national leadership and the VA HIV/Hepatitis Quality Enhancement Research Initiative.

 Although there have been reports of telemedicine care for HIV and hepatitis C in non-VA settings,

 we are unaware of published reports describing HIV/hepatitis C telemedicine efforts within the VA system. To address this clinical need, the VA Greater Los Angeles Health System (VAGLAHS) implemented HIV and hepatitis C telemedicine services at 2 rural clinics.

The goal of this study was to assess the impact of telemedicine services for the VAGLAHS HIV and hepatitis C clinics on patient-oriented outcomes. We used clinic completion rates as a proxy measure for access. Our primary hypothesis was that telemedicine would safely improve clinic completion rates. We also hypothesized that telemedicine would be associated with high patient satisfaction and reduced health visit—related time.

This is a pre-post intervention study of HIV/hepatitis C clinic visits clustered by patients. We used the pre-intervention, in-person clinic completion rate of each patient as an internal control group to assess the impact of the telemedicine clinics. The VAGLAHS HIV clinic initiated telemedicine consultations in February 2009 and the VAGLAHS hepatitis C clinic began telemedicine consultations in February 2010. Pre- and post-intervention data for this study were collected from June 2008 to June 2011. This study was approved by the VAGLAHS Institutional Review Board.

VAGLAHS has an existing telemedicine infrastructure that includes “hub” sites located in urban Los Angeles. The “spoke” sites in this study included the Santa Maria and Bakersfield clinics, which are respectively 160 and 110 miles driving distance from the urban clinics. Telemedicine consultations are performed utilizing video teleconferencing end units over a high-speed Internet connection (Polycom; Pleasanton, California). At the hub sites, mobile end units with 32" monitors can be moved between clinics. A clinical manager at each of the hub and spoke sites is responsible for providing support to the telemedicine program. The HIV/ hepatitis C telemedicine clinics utilized this existing infrastructure, and no additional funds, staff, or equipment were required for implementation. However, the “clinical champions”

and nursing staff for these clinics did provide in-kind time to develop policies and processes to implement telemedicine services.

Telemedicine consultations were performed by healthcare providers during the existing infectious disease and hepatitis C clinic times at the urban hub sites. For both the HIV and hepatitis C clinics, there was a single half-day clinic per week. Patients scheduled for a telemedicine consultation presented to a spoke site. An onsite nurse measured vital signs, and patients were placed in a private examination room set up with teleconferencing equipment. The healthcare providers at the hub sites could remotely conduct consultations; perform interviews and patient education; view chart notes, radiology tests, and electrocardiograms; enter orders for pharmacy,

laboratory, and radiology; and perform documentation using the VA computerized patient record system. During the study period, there was a single provider in each of the HIV and hepatitis C clinics who performed all telemedicine consultations (CJG, JPS).

The study population is a convenience sample of patients residing in the Santa Maria and Bakersfi eld catchment areas who were referred by their HIV or hepatitis C provider to the telemedicine clinic. Providers selected patients who were deemed to have stable disease. There were no other eligibility criteria such as viral load or active antiviral therapy.

The primary binary outcome was clinic completion, defi ned as the completion of a scheduled appointment. Reasons for non-completion of a scheduled appointment included no-shows and patient cancellations. Cancellations by the clinic were excluded from this analysis.

Determination of clinic completion was performed by review of electronic medical records and administrative data. All telemedicine clinic encounters were reviewed. To create patient-specifi c internal control groups, we also reviewed all condition-specifi c in-person visits that occurred during the study period. For example, if a patient was scheduled for an HIV telemedicine clinic visit, we collected data on all prior scheduled HIV in-person clinic visits for that patient.

Secondary outcomes included patient satisfaction and health visit—related time. At the end of each telemedicine consult, a research assistant administered a 10-minute survey from the hub site. Overall satisfaction and ratings of physician interaction, privacy, physical environment, and convenience were measured on a 5-level Likert scale (very dissatisfied, somewhat dissatisfied, neither satisfi ed nor dissatisfied, somewhat satisfied, very satisfied). Preference for in-person versus telemedicine consultation was measured on a binary scale. Patients were also asked to estimate travel and on-site time required for in-person and telemedicine visits. For estimates of on-site time, patients were asked to consider time for checking in, waiting, and completing the consultation. Total health visit–related time is defi ned as the sum of travel and on-site time related to a clinic visit.

We collected demographic data on age at time of the first telemedicine visit, gender, and race/ethnicity from review of electronic medical records and administrative data. Race/ethnicity data were self-reported to VAGLAHS. Finally, we assessed whether there were any hospitalizations at a VA facility or deaths that occurred within 30 days of a subspecialty clinic.

All electronic and administrative records were reviewed by a research assistant (HNS). To assess inter-rater reliability, a physician-investigator (BCS) blinded to the research assistant’s abstractions independently re-reviewed notes from 50 randomly selected visits. There was 100% concordance on all data elements, including visit completion, demographic data, and occurrence of 30-day hospitalization or death.

We performed unadjusted and adjusted analyses of the primary outcome of clinic completion rates. Contingency table methods were used to estimate unadjusted clinic completion rates for scheduled in-person and telemedicine visits. To assess the adjusted effect of telemedicine on clinic completion, we estimated a logistic regression model with personfi xed effects. The binary outcome is clinic completion, and the predictor is clinic setting (in-person vs telemedicine). This approach controls for the clustering of visits by patient, uses each patient’s in-person clinic experience as an internal control group, and eliminates confounding by person-level factors. Odds ratio and 95% confi dence intervals were calculated directly from the regression results. We performed sensitivity analyses with person-level random effects, and with no person-level effects; there were no qualitative differences in the results.

Although odds ratios are commonly reported effect measures, they may be difficult to interpret clinically. We used a bootstrapping approach to estimate the adjusted impact of telemedicine on clinic completion rates. We generated 1000 re-samples of the patient population of equivalent size to the original patient population. For each sample, we fit the logistic regression model with patient-level fixed effects, and the absolute rate difference was estimated by hypothetically setting all visits to either in-person or to telemedicine. We used estimates across all 1000 re-samples to generate a mean point estimate and 95% confi dence intervals of the absolute rate difference between in-person and telemedicine visits.

To explore the possibility of condition-specific effects of telemedicine on visit completion, we re-fit the regression model with an interaction term between telemedicine and condition (HIV vs hepatitis C).

We describe all secondary outcomes in descriptive tables. All time-interval outcomes were highly right-skewed. Using patient estimates of health visit—related time, we calculated the median and interquartile range of the differences between in-person and telemedicine clinic visits. We used the Wilcoxon rank-sum test for non-parametric data to assess time-based measures.

Data entry was performed using Microsoft Access (Redmond, Washington). Data management and analyses were performed using SAS 9.1 (Cary, North Carolina).

There were 43 patients who accounted for 94 telemedicine visits and 128 in-person visits to HIV or hepatitis C subspecialty clinics. This cohort was predominantly male, as expected for a VA population, and the majority of patients were evaluated for hepatitis C (

The in-person clinic completion rate of the study cohort was 61%, which is similar to the historic in-person clinic completion rate of 61% for HIV and hepatitis C visits by all patients from the study catchment areas from 2008 to 2010 (

= 0.9). The adjusted odds ratio of clinic completion associated with telemedicine was 2.3 (95% CI: 1.0-4.7; 

 = 0.04). In bootstrapped analysis, telemedicine was associated with an increased clinic completion rate of 13% (95% CI: 12-13) (

). There was no evidence of a statistically important interaction effect (

 = 0.2) between telemedicine and condition (HIV vs hepatitis C).

Of the 43 study cohort patients, 30 (70%) completed a telemedicine-facilitated survey. The remaining 13 patients did not complete the survey either because they did not show for their telemedicine visit or because of research assistant unavailability. There were no refusals to participate in the survey. Patients were highly enthusiastic about telemedicine, and >95% of respondents gave the highest possible rating for overall satisfaction and expressed a preference for telemedicine over in-person clinic visits (

Respondents estimated that total health visit time was 340 minutes less for telemedicine compared with in-person visits (

 <0.001); time reduction was attributable to both shorter travel and on-site times (

). However, the majority of perceived time reduction was related to travel. Approximately 65% of respondents reported that they traveled to the urban hub site by car, and the remainder used a VA shuttle or public transportation.

CLINICAL

In an integrated health system, human immunodeficiency virus and hepatitis C telemedicine clinics are associated with improved access, high patient satisfaction, and reduction in health visitâ€œrelated time.

Evaluation of Human Immunodeficiency Virus and Hepatitis C Telemedicine Clinics

To test the usefulness of the Diabetes Complications Severity Index (DCSI) without laboratory test results in predicting healthcare costs, for potential use in disease management programs.

Retrospective cohort study using up to 2 years of claims data from 7 health insurance plans.

Individuals with diabetes mellitus and continuous enrollment were study subjects. The DCSI (sum of 7 diabetes complications graded by severity as 0, 1, or 2; range 0-13) and count of diabetes complications (sum of 7 diabetes complications without severity grading; range 0-7) were the main independent variables and were generated using only diagnostic codes. We analyzed 5 types of healthcare costs (ie, total costs, inpatient costs, hospital other costs, pharmacy costs, and professional costs) attributable to the DCSI and the complication count with linear regression models, both concurrently and prospectively.

The DCSI without laboratory data was a better predictor of costs than was complication count (adjusted R2 of total costs: 0.095 vs 0.080). The DCSI explained concurrent costs better than future costs (adjusted R2 of total costs: 0.095 vs 0.019). There were important differences in healthcare utilization among people stratifi ed by DCSI scores: 5-fold and 3-fold differences in concurrent and prospective total costs, respectively, across 4 DCSI groups.

The DCSI without laboratory data may be useful for stratifying individuals with diabetes into morbidity groups, which can be used for selection into disease management programs or for matching in observational research.

The Diabetes Complications Severity Index (DCSI) without laboratory test results can be used to explain 5 different types of healthcare costs: total costs, inpatient costs, hospital other costs, pharmacy costs, and professional costs.

		 Significant differences in medical utilization existed among people stratifi ed by DCSI scores without laboratory test results.

		 The DCSI without laboratory test results may be useful for stratifying individuals with diabetes into morbidity groups as a selection criterion for disease management programs or as a matching factor for research.

The increasing prevalence and growing medical expenditures for diabetes mellitus have made it one of the major healthcare issues in the United States.

One approach to containing diabetes-related medical expenditures is to enroll diabetes patients in disease management programs so that their medical care can be integrated while their quality of medical services is monitored.

 However, due to limited resources, a healthcare plan cannot enroll all patients in such programs. Because a very small proportion of patients consume a relatively large amount of medical resources,

 targeting disease management to these patients may be an effi cient use of resources.

The Diabetes Complications Severity Index (DCSI), developed by Young and colleagues,

 uses diagnosed complications and laboratory results to assess the level of risk of adverse outcomes for diabetes patients, including hospitalizations and mortality. Even though the DCSI score has just been established, it has been quickly adopted by many researchers.

 However, DCSI’s utility as a risk measure for individuals with diabetes may be limited. First, laboratory test results were included during the construction of the DCSI, limiting its use if laboratory results are unavailable, as is typical in most administrative data. Second, the performance of the DCSI has not been evaluated for explaining costs—an outcome that is important for healthcare providers and health services researchers. Third, the DCSI was developed using a patient sample from Washington State; its generalizability to other populations with diabetes is untested.

Therefore, we aimed to extend the usefulness of the DCSI without laboratory test results as a predictor of diabetes severity and to consider its possible use in disease management programs. We previously compared the DCSI with laboratory results with the DCSI without laboratory results for predicting hospitalization, and found that the index was similarly predictive with and without laboratory values.

In the current study, we addressed 3 hypotheses. First, the DCSI without laboratory results is a good predictor of healthcare costs. Second, the DCSI without laboratory results is a better measure of diabetes severity than counts of complications. Third, the DCSI without laboratory results explains concurrent costs better than future costs.

This was a retrospective cohort study using up to 2 years of claims data in which we tested the value of the DCSI for explaining medical utilization and classifying individuals with diabetes without laboratory data.

We accessed claims data from 7 Blue Cross Blue Shield plans; the detailed information was described previously.

 Th e original data, collected from 2002 to 2005, were subsequently updated with additional data on the original individuals through December 2006. The following data were acquired: (1) enrollment fi les for administrative data; (2) benefi ts information to determine medical and pharmacy coverage; and (3) inpatient, outpatient, and pharmacy claims records containing 

International Classifi cation of Diseases, Ninth Revision, Clinical Modification

) diagnosis, Current Procedural Terminology codes, National Drug Codes, costs, and charges (submitted, allowed, and paid).

For inclusion in our analytic cohort, we required that the enrollees have diabetes and 12 months of coverage, including pharmacy coverage, in each calendar year over the study period.

We defined individuals as having type 2 diabetes mellitus if they had 1 relevant inpatient code or 2 outpatient

 codes separated by at least 30 days. The relevant codes were 250.xx, 648.0 (diabetes mellitus with pregnancy), 362.0 (diabetic retinopathy), or 266.41 (diabetic cataract). Individuals with only code 250.x3 (type 1 diabetes mellitus) were not included. Additionally, any individual fi lling a prescription for a medication for treatment of hyperglycemia was included (

). Combination medications were also identifi ed. If the prescription was for metformin alone, the individual was also required to have an

code for diabetes for inclusion in this group. The calendar year of the earliest diagnosis of diabetes was used as the starting point of the observation period.

Costs were obtained from the claims. Total costs were examined as well as several subcategories: inpatient costs, hospital other costs, pharmacy costs, and professional costs. These costs were standardized to adjust for differences in reimbursement between plans and for infl ation over time. Professional services refer to services submitted on professional claims, and include things such as visit costs, laboratory costs, and diagnostic imaging.

Inpatient costs were standardized using the Diagnosis-Related Groups (DRGs) and the Centers for Medicare & Medicaid Services (CMS) national average inpatient cost per DRG. Professional services costs were standardized using the procedure code and the CMS Resource-Based Relative Value Scale unit for the procedure. All other costs were calculated using the total amount paid as submitted on the claim (total amount paid = amount paid by plan + amount paid out of pocket by enrollees including copay, deductible, and coinsurance + amount paid for coordination of benefi ts). If an inpatient cost or professional services cost could not be standardized due to an invalid DRG or invalid procedure code, respectively, it was standardized using the total amount paid.

DCSI, DCSI Complication Count, and Covariates

To replicate the DCSI and DCSI complication counts, we used 

codes and the classifi cation method described by Young and colleagues.

 The DCSI consists of scores (0, 1, or 2) from 7 complication categories: retinopathy, nephropathy, neuropathy, cerebrovascular, cardiovascular, peripheral vascular disease, and metabolic (range 0-13). The DCSI complication count is a count of any complication in the 7 categories (range 0-7). We did not include laboratory results in constructing DCSI and DCSI complication counts. Inclusion of laboratory results would only affect the complication category of nephropathy. Whereas 6 diagnosis codes could contribute 1 to the DCSI score, 3 laboratory results could have the same contribution; in addition, whereas 3 diagnosis codes could contribute 2 to the DCSI score, only 1 laboratory result could make the same contribution.

We performed 2 sets of analyses in this study: concurrent costs and prospective (next-year) costs. The complete enrollment requirements for the concurrent and prospective analyses were 1 year and 2 years, respectively. For both analyses, the information to construct the independent variables came from the calendar year of the qualifying diabetes diagnosis, to match what was done by Young and colleagues.

 Outcome variables were derived from the calendar year for the concurrent analysis and the subsequent year for the prospective analysis.

We applied a linear regression model for estimating costs because it generates the highest adjusted R

We reported and compared the incremental increase of costs and measures of overall model fit, the adjusted R

 for costs was obtained from split-half analysis to control for overfitting. We tested inclusion of the main independent variables, DCSI and DCSI complication count, categorically (0, 1, 2, 3, 4, 5+) and linearly. In categorical analysis, the incremental increase of costs was derived by comparing samples in a given category with those in category 0; in linear analysis, it was the incremental increase of costs associated with 1 unit of increase in either the DCSI or DCSI complication count.

Use of DCSIs for Disease Management Programs

We divided the study sample into 4 groups by DCSI: 0, 1, 2, and 3+, and compared characteristics by DCSI groups.

The data were de-identifi ed in accordance with the Health Insurance Portability and Accountability Acts (HIPAA) defi nition of a limited data set. The Johns Hopkins University Office of Research Subjects deemed the study to be exempt from federal regulations because the research activities were considered to be of minimal risk to subjects, as the subjects were not identifi able.

METHODS

The Diabetes Complications Severity Index without laboratory test results is a good measure of diabetes severity, given its ability to explain costs.

Predicting Costs With Diabetes Complications Severity Index in Claims Data

The emergence of information technology in healthcare holds the promise to transform the industry through the creation of highly reliable information exchange. These same technologies have a central role in the patient safety movement. Organizations that wish to deliver safe and high-quality healthcare will only be successful if they plan, develop, and use health information systems with the principles of high-performing organizations in mind. We discuss the current state of health information technology in the patient safety movement, how this technology can contribute to high organizational performance, and some caveats.

Healthcare organizations can advance patient safety by using the principles of high-performing organizations when they design and implement health information technology tools. Key principles include:

		 Engaging healthcare workers to design and provide feedback on health informatics tools.

		 Promoting electronic error-reporting systems as a vehicle to identify errors and swarm and solve problems in real time.

		 Leveraging technology such as computer screen savers and handheld electronics to present safety data to healthcare workers in the course of their daily work.

		 Educating all healthcare workers in health information technology; developing and promoting a cadre of experts in the field to meet the growing needs of our healthcare industry.

Despite increasing awareness of the risks to patients within the US healthcare system and the human and economic toll that medical errors exact, improvements in patient safety have been slow.

 While individual accountability for patient outcomes is essential for the delivery of high-quality medical care, the complexity of healthcare requires a highly reliable system built within a culture that recognizes errors and process defects as opportunities to learn and continuously improve. The Institute of Medicine report

 emphasized the important role of health information technology in preventing harm to patients.

 Through the automation of errorprone tasks and decision support systems designed to minimize reliance on human memory, health information technology has rapidly become an important tool to address problems faced by healthcare organizations and their patient safety programs.

This commentary reflects how high-performing healthcare organizations use health information technology to advance patient safety.

April 2012

Factors of Hyperlipidemia Medication Adherence in a Nationwide Health Plan

Predictors of Physician Use of Inpatient Electronic Health Records

Evaluation of Human Immunodeficiency Virus and Hepatitis C Telemedicine Clinics

Predicting Costs With Diabetes Complications Severity Index in Claims Data

Chasing High Performance: Best Business Practices for Using Health Information Technology to Advance Patient Safety

Association Between Physician Quality Improvement Incentives and Ambulatory Quality Measures

Determinants of Diagnostic Imaging Utilization in Primary Care

Cost-Effectiveness of Pharmacist-Provided Treatment of Adult Pharyngitis

Primary Care Physicians' Experiences With Case Finding for Practice-Based Care Management