Improving Diabetes Care in the Clinic Jayendra H Shah
Chapter Notes

Save Clear

Diabetes Registries—A Practical Guide to Design, Selection, and Its Use in the Management of DiabetesChapter 1

Glen H Murata


Most major industries—retailing, manufacturing, transportation, financial, hospitality, energy, and natural resources, etc.—rely on information technology to perform a systematic evaluation of operations, identify where processes are suboptimal, and triage those problems to appropriate personnel for corrective action. No one would board an airplane if the pilot not only had to fly the plane but chart the course on paper using a sextant and compass. Moreover, the chances of recovery would be small if the first indication of trouble is a sudden loss in altitude. However, our health care industry has been operating in this mode for decades—without a clear indication of what works and what doesn't, which patients need care and which don't, and which providers are doing well and which are not. Moreover, the first indication of trouble is an adverse patient outcome for which there may be no recovery, exhaustion of resources resulting in withdrawal of care, or poor job satisfaction leading to a decline in the physician workforce. Fortunately, health care systems are now adopting information systems which will bring their enterprises into the modern age. Because they focus on the management of entire populations, registries are the most promising of many tools that will drive much-needed and much-delayed health care reform. As one of the most complicated chronic diseases, diabetes should be the focus of one of the first registries at almost every institution.
Webster's online dictionary defines “disease registry” as a “system of ongoing registration of all cases of a particular disease or health condition in a defined population” ( Although there is a tremendous variation in their functionality, registries share common features including disease focus, specified purpose, targeted population, sponsorship, data 2sources, data elements, storage format, architecture, data entry, analytical techniques, and reporting services. Each of these domains has undergone a remarkable evolution in recent years. The number of diseases or conditions tracked by disease registries has increased substantially in countries around the world.1 Registries are now used to track the incidence, prevalence, and risk factors for disease, often to mitigate exposures in populations; to track the natural history of disease; to assemble enough rare cases for systematic study; for research on the association between treatments and outcomes; for quality improvement (QI) through profiling and identifying best practices; for health policy and financial planning; to manage populations by characterizing the needs of and prioritizing services for its members; and to support decision making for individual patients. The targeted population can vary by age, gender, ethnicity, geographic locale, or entitlement. Although registries were first developed by government agencies for epidemiologic purposes, they are now sponsored by professional organizations, health care systems, insurance companies, and pharmaceutical firms. The earliest registries were populated by case reporting from individual clinicians; now, many of the most useful retrieve data automatically from electronic medical records. Data elements range from coded or standardized data to free text and even images. Registries that used to be recorded on paper are now stored in large relational databases residing on institutional computers. Sophisticated software can derive new clinical parameters from raw data and perform sophisticated analyses well beyond the reach of most clinicians. Finally, results can be reported to users through the internet or reporting services of many database software applications.
One of the most common applications for these registries is to help institutions, practices, or individual clinicians deal with chronic diseases. Managing a chronic illness often requires follow-up over years, multiple visits, intensive and repeated testing, many consultants, and care delivered in different settings. Problems may occur with transitions of care from provider to provider, from setting to setting, or from one phase of the disease to another. As a result, their treatment tends to become fragmented, delayed, or incomplete over time resulting in suboptimal outcomes, poor quality of life, and diminished patient satisfaction. Chronic diseases are particularly vulnerable to problems related to access, financing, work force, distribution of services, and legal liabilities. Because most health care systems are struggling with these issues, it is no surprise that there is a surging interest in a technological solution to the problems.
The health care crisis in the United States and many other countries represents the confluence of several social trends—poor access to care, dismal health outcomes, unsustainable costs, patient and provider dissatisfaction, and erosion of the primary care base. To a large extent, this “perfect storm” is due to a delivery system based 3upon the “acute care model”. In this paradigm, encounters are initiated by the patient in response to symptoms. This strategy often places the onus for first contact on the person who has the least knowledge about his/her condition or the illness in question. For example, a patient with colonic polyps can miss the next colonoscopy because he or she was unaware of the pathology results, its implications, or recommendations for surveillance. Planning by the physician is done on a case-by-case basis so that it is not possible to anticipate the needs of entire populations or set priorities to maximize collective well-being. Finally, the intervention is reactive, i.e., to ameliorate a condition that has progressed beyond the preclinical phase. The opportunity to treat the condition at an earlier and often more responsive stage is lost.
These problems are compounded by reliance upon the outpatient visit as the principal means of delivering medical services. The “visit-based” approach precludes the neediest patients—those with access barriers who never present for treatment. Progress is tied to the next available appointment, not to the responsiveness of the disease to treatment. For example, insulin titrations often occur every several months even though treatment response can be assessed in just a few days. Despite its disadvantages, administrative processes and financial incentives still favor the acute care, visit-based approach. Examples include the need to identify a “primary diagnosis” for an outpatient encounter and higher reimbursement for an office visit than an equally effective telephone call. The more efficient approach is to track the health status of an entire population, focus attention on the highest priorities, recall patients when services are needed as well as requested, intercede at early rather than late stages of illness, and triage patients to team members with the most appropriate skill level.
Meaningful health care reform, i.e., changes to improve quality while reducing costs, requires a fundamental re-engineering of the nation's health care system. “Population management” provides the overarching conceptual framework in which such reform can occur. Its components include greater emphasis on collective performance, societal impact, and affordability; more realistic expectations of and more active participation by patients; increased accountability of providers for reporting outcomes and justifying expenditure of common resources; prioritization based upon need and likelihood of benefit; optimization based upon correlating processes and outcomes; reducing disparities due to race/ethnicity, geography, or economic status; and setting standards for the delivery of services through health policy. In this new paradigm, health care institutions would develop guidelines that are data driven, evidence based, and individualized to each patient's circumstance; enhance the value of services through performance reviews and cost analyses; provide seamless care through different settings and phases of illness; allocate resources based upon realized and unmet demand for services; assess performance on a 100% sample of cases instead of random audits; and charge QI personnel with interventional as well as analytical responsibilities. Clinicians should also practice in a different way. The emphasis would shift from processes to outcomes; the setting 4would shift from the office to anything that works; prevention would take precedence over cure; consultations would be prioritized by patient need instead of provider preferences; continuing medical education (CME) would be driven by provider performance instead of interest; practice guidelines would be replaced by individual risk stratification; and equal attention would be given to the psychological, social, and biological determinants of outcomes. All of these needed reforms depend upon gathering comprehensive data on the personal, provider, practice, health system, and national level.
Fortunately, the electronic medical record provides an unprecedented opportunity for promoting these reforms. “Front-end” capabilities include shared access to a comprehensive medical record; practice support as exemplified by automated forms processing, computerized order entry, and treatment advisories; enhanced patient safety through alerts and warnings; and decision-support based upon complex processing of clinical data beyond the reach of a busy clinician. However, it is the “back-end” functionalities that will enable providers, health systems, and government agencies to manage resources in a way that optimize the health status of populations. Systems can now tap into the stream of data populating clinical data repositories. These innovations include cohort tracking in which the work-up of patients is monitored in real time; visit management in which patients are “prepped” for an upcoming visit; prevention management in which services are coordinated at an institutional level on behalf of clinicians; pharmacy management in which indications, safety, cost, effectiveness, and adherence are concurrently monitored; and consult management in which appropriate cases are referred to specialists even if they are “missed” by their primary care providers (PCPs). Registries are the most influential of these back-end functionalities. They can track the status of an entire population with a given condition and stratify its members by risk, need, severity, or complexity. Identifying patients at high risk allows providers to focus behavioral modification and preventive services on the group most likely to benefit. Some patients may have greater need for services than others because they are less able to afford such care on their own, more likely to respond, suffer greater consequences of unmanaged illness, or are more overdue. Those with more severe disease should be given higher levels of care than milder cases. Finally, case management and care coordination would be reserved for complex cases involving multiple specialties.
Diabetes is one of the most common diseases for which care coordination is critical. Patients must adhere to a healthy diet, exercise regularly, watch their weight, take (and even adjust) their medications regularly, monitor their own blood sugars, and care for 5their feet. Clinicians must provide eye screening and foot care; give vaccinations; offer counseling on diet, exercise, and weight loss; manage hyperglycemia to reduce the risk of microvascular complications; treat hypertension and hypercholesterolemia to decrease the incidence of macrovascular events; manage devastating complications such as blindness, end-stage renal disease, myocardial infarction, or stroke; and even deal with the psychological and social consequences of the illness. The latter include depression, loss of employment, economic deprivation, and family discord related to the need for care giving or change in eating habits for the entire family. It is not surprising that diabetes was the focus of the earliest disease registries.24
It is beyond the scope of this chapter to describe existing diabetes registries in detail. Besides, many are proprietary and their functionalities are offered only to members who subscribe. Suffice it to say that long-standing registries have been sponsored by municipalities (New York City), academic health centers (Penn State Milton S Hershey Medical Center), specialty care centers (Joslin Diabetes Clinic), integrated health systems (Kaiser Permanente), programs in foreign countries (National Diabetes Surveillance System in Canada), and international collaborations (DIABCARE Q-NET in Europe).
Several studies have analyzed the impact of diabetes registries on the quality of patient care.510 Although a variety of approaches has been used, most have demonstrated a favorable effect. Studies of this type can be very difficult to design, conduct, and analyze. Registries can identify patients who need critical services, but an improvement in outcomes requires an effective action plan for the findings. This plan can fail because of “clinical inertia”, resistance to change, hostility over fault finding or loss of autonomy, confusion about roles and responsibilities, and suspicions about the quality of the data. Even if clinicians are motivated, corrective actions may require additional personnel, staff training, assigning new tasks, changing the practice's priorities, and finding the time and expertise for managing the data. Perhaps the most important impediment is that registry-driven care may not generate revenue and is therefore given lower priority than activities that do. Thus, the benefits of a registry may depend more upon the practice's action plan than on the quality of information presented. In addition, their full impact may not be realized until reimbursement is aligned with patient outcomes instead of the workload generated by conventional activities.
Another problem is that the “gold standard” for efficacy—the randomized clinical trial (RCT) is not feasible for diabetes registries. The RCT is considered the best test of an intervention because it is the only method that handles measured and unmeasured patient covariates, i.e., patient attributes that affect the study's end-point other than the groups to which subjects have been assigned. Randomization balances the prognostic factors across the intervention and control groups whether they are measured or not, 6while statistical methods can only handle those that are measured. Many patient attributes (such as motivation, family support, or commitments to a job or school) are important determinants of diabetes outcomes but not measurable. RCTs are difficult to apply to diabetes registries for practical and ethical reasons. Patients might object to being randomly assigned to practices with or without registries. On the other hand, it is unreasonable to expect practitioners to consult a registry for intervention patients but not for controls when information is available for both. Finally, trials are ethical only if there is “equipoise” or legitimate uncertainty about the risks and benefits of the two treatment arms. It would be difficult to argue that knowing about a critical deficiency in patient care can either be better or worse than not knowing about it.
As a result, most studies testing the benefits of diabetes registries resorted to less rigorous study designs—the pre/post study, the cohort study, and the group-randomized trial. A pre/post study compares outcomes before and after the implementation of a registry. For example, Pollard et al.6 studied processes and outcomes of diabetes care in 661 subjects at six federally-qualified health care centers in West Virginia. Data pre- and post-registry were compared for three levels of registry utilization. The registry significantly improved 12 of 13 process and three of six outcomes measures for patients exposed to at least medium levels of registry utilization. Likewise, Ciemens and associates9 analyzed metabolic status and preventive services in 495 adult diabetic patients through a baseline and two intervention phases. The latter consisted of a “low-dose” phase emphasizing provider and patient education and a “high-dose” phase based upon the registry and workflow changes. Significant improvements were noted in blood pressure, glycosylated hemoglobin, low-density lipoprotein, and the proportion of patients receiving recommended eye, foot, and renal evaluations.
Pre/post studies should be interpreted with caution if there is temporal drift in the study's end-points, co-interventions have been introduced, efforts to measure end-points in the “pre” phase are not robust, and outcomes are sensitive to patient or provider motivation. Temporal changes in the study's outcome may offset the benefits of registries—giving the impression that they are ineffective. For example, suppose that the prevalence of childhood obesity is steadily increasing in a community. A registry that tracks patient status and guides interventions may stabilize the incidence. However, a pre/post analysis would show no changes. On the other hand, a spontaneous decline in the rate could be attributed to the registry when, in fact, it had no effect. Co-interventions are those introduced into practice while a study is underway. One example is the patient-centered medical home. If the co-intervention is adopted during the “post” phase, it would not be possible to separate its effect from the use of a registry. In many pre/post studies, outcomes during the “pre” phase are measured by routine procedures, while in the “post” phase, they are evaluated in a prospective and robust manner. For end-points that are silent or have atypical presentations, there can be an apparent change from “pre” to “post” that represents a measurement artifact. For example, myocardial infarction can be “silent” among 7diabetic patients. In the “pre” phase they might be missed, while in the “post” phase they might be picked up by electrocardiogram (EKG) surveillance. Finally, because so many diabetes outcomes depend upon patient or provider motivation, a spurious benefit of registries might arise from the “Hawthorne” effect. This effect is a change in a study's outcome due to participants' knowledge that they are being observed. Improvement might be attributed to registries when it actually represents a more conscientious effort by providers or greater adherence to recommendations by patients. Thus, study designs using parallel controls tend to yield more convincing evidence of efficacy than pre/post studies.
A cohort study compares two or more groups that differ by an important attribute. However, the groups are assembled without specific techniques (such as randomization, matching, or stratification) to assure that they are similar. This design is exemplified by the study of Coppell and coworkers.7 These investigators compared 3,646 patients enrolled to the Otago Diabetes Registry in New Zealand with 1,103 who were not. Enrolled patients were more likely to receive recommended process measures as well as angiotensin-converting enzyme (ACE) inhibitors, other antihypertensive medications, and lipid-lowering agents. However, because randomization was not used, the groups could have been imbalanced with respect to patient characteristics affecting completion of recommended tasks. Moreover, because the practices were not randomized to registry-driven versus routine care, the apparent benefit of registries could have been due to differences in office policies or procedures other than the registry. This bias is possible because practices that adopt registries are likely to use other measures to assure high quality of care.
Randomizing practices to the two study arms is a better approach than a cohort design because it addresses such differences. This approach is known as a group-randomized trial. The terminology refers to the fact that patients belonging to each practice are assigned en bloc to the arm to which the practice has been randomized. As noted above, the principal advantage is that covariates arising from the practice level are balanced. Nevertheless, special statistical procedures are needed to adjust for differences in patient attributes as well as cluster effects. The latter result from the fact that the data are “nested” (i.e., patients are grouped under practices). It is beyond the scope of this chapter to discuss the implications of this data structure other than to note that the power of the study is reduced if there is significant variation in outcomes across the practices. Fischer et al.8 used a sophisticated design to test whether patient and provider report cards generated from a computerized diabetes registry had an effect on glycemic, lipid, and blood pressure outcomes. Randomization for mailed patient report cards was done at the patient level. However, a 2 × 2 factorial design was used to evaluate the effectiveness of point-of-care patient report cards and provider report cards containing patient-specific data. The four arms in this design were: patient + provider report cards, provider report cards alone, patient report cards alone, and neither. One large and one small practice were randomized to each of these four arms. Neither the mailed nor point-of-care patient report cards had a beneficial effect. 8Patients of providers receiving report cards were more likely to achieve their targets for glycemic control than those of control providers (6.4% versus 3.8%, respectively; P < 0.001). This study represents the most rigorous evaluation of registry-generated reports to date, although the benefits were modest at best. Unfortunately, treatment plans were not standardized, and it was not clear that the practices had resources to act upon the information.
In summary, most studies have shown that diabetes registries improve the quality of care, although the issue is by no means settled.11,12 The evidence supporting this conclusion is modest because of limitations in study design or in the plans for corrective actions. Finally, as pointed out by Trief and Ellison,12 diabetes registries are not without risk. Harm could arise from the intrusion of a third party (the registry developers) into an otherwise confidential patient-provider relationship as well as loss of protected health information.
Like others,13,14 we will focus our remaining discussion on developing population management tools at the New Mexico Veterans Affairs Health Care System (NMVAHCS). The objective is to review the features of registries that the reader may wish to consider when selecting or designing one for his or her own situation.
The first functionality of a diabetes registry is to identify appropriate patients. Many factors will affect case identification including its sensitivity and specificity, data sources, search criteria, data validity, search period, sampling biases, and the number of criteria required.
Sensitivity and Specificity
Registries that are highly sensitive capture most patients who have diabetes at the expense of misidentifying some who do not. On the other hand, those that are specific will correctly classify cases that are retrieved but miss cases with marginal criteria. There are no optimal criteria that serve every purpose. For example, if the purpose of the registry is to prevent the progression of early diabetes, then highly sensitive criteria should be used. However, if the purpose is to recruit patients with advanced disease to a clinical trial of experimental treatment, then highly specific criteria are more appropriate. At the NMVAHCS, we build the most highly sensitive and comprehensive registries possible. Our reasoning is that subsets can then be selected by querying the master registry. This approach is more efficient than building a different registry for every purpose.
Data Sources
Diabetes can be identified from five conventional data sources: (1) hospitalization records, (2) outpatient encounters, (3) problem lists, (4) laboratory tests, and (5) medications.
TABLE 1   Data Sources and Case Identification (n = 13,999)
Cases identified
First criterion met
Single criterion*
3,498 (25.0%)
553 (4.0%)
216 (6.9%)
HbA1c ≥6.5%
10,134 (72.4%)
5,146 (36.8%)
1,070 (34.4%)
8,647 (61.8%)
1,382 (9.9%)
80 (2.6%)
Problem lists
11,165 (79.8%)
4,095 (29.3%)
908 (29.2%)
10,780 (77.0%)
2,823 (20.2%)
839 (27.0%)
*3,113 were diagnosed by only one criterion; the number in parentheses is the proportion of these cases detected by each of the five criteria.
On occasion, other files can be interrogated as well. For example, procedure files can be searched for laser photocoagulation—an intervention most commonly used for diabetic patients with proliferative retinopathy. Likewise, pathology files might be used to identify patients with nodular glomerulosclerosis (Kimmelstiel-Wilson disease). However, caution should be exercised when using these unconventional data sources because the etiology might not be specified. For example, neither laser treatment nor Kimmelstiel-Wilson disease is entirely specific for diabetes. Table 1 shows the data sources used to identify 13,999 patients at NMVAHCS in 2011.
Note that there are marked differences in the proportion of cases identified depending upon the source. Problem lists capture the greatest proportion of patients followed by outpatient encounters and hemoglobin A1c (HbA1c). However, HbA1c is most frequently the first criterion met and is the most common finding in 22% of patients with a single criterion.
Search Criteria
The search for cases is greatly facilitated if data are coded or expressed in standard terminology. Commonly used coding systems include the International Classification of Diseases (ICD), e.g., ICD-9 for diagnoses, Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT) for pathology specimens, Current Procedural Terminology (CPT), e.g., CPT-4 for interventions, and Logical Observation Identifiers Names and Codes (LOINC) for laboratory tests. Many of these codes have several dimensions so that the search can be highly refined. For example, LOINC has six parts: (1) name, standardization, and type of challenge (if any); (2) the physical property; (3) the timing of measurements; (4) the organ system source and type of specimen; (5) the scale of measurement; and (6) the analytical method. Searching for nonstandard terms is a challenge because clinicians may describe a disease in a wide variety of ways. For example, a radiologist may use the terms “pneumonia”, “pneumonitis”, “infiltrate”, “alveolar filling pattern”, “consolidation”, “haziness”, or “air bronchograms” to describe the same abnormality on a chest X-ray. The problem is 10amplified if the term of interest is not placed in a dedicated field but rather buried in free text. Not only must the term be found, but the user must deal with qualifiers such as “probable”, “possible”, “consistent with”, “not likely to be”, or “rule out”. While extremely valuable, free-text searching adds a whole new level of complexity to searching. For this reason, our diabetes registry is built only on coded data and standardized terminology.
Discharge diagnoses, outpatient encounters, and problem lists are usually coded using ICD-9 or ICD-10. The Appendix lists the ICD-9 codes used for diabetes at NMVAHCS. Note that, because our philosophy is to use highly sensitive criteria, the Appendix includes gestational and neonatal diabetes, diabetes complications (e.g., ICD-9 code 364.42: rubeosis iridis), laboratory abnormalities [ICD-9 code 790.22: impaired glucose tolerance test (oral)], and even adverse side effects or poisoning by diabetic agents (ICD-9 codes 962.3 and 932.3). Including terms used by specialists is important for tertiary facilities. Many PCPs in the community might use the facility only for consultative purposes. In that case, terms most commonly used for diabetes (e.g., ICD-9 code 250.00: diabetes mellitus without mention of complication, type II or unspecified type, not stated as controlled) will not be found in outpatient encounters or problem lists. Including symptoms, signs, and laboratory tests may capture patients for whom the diagnosis is likely but not firmly established. The rationale for including adverse drug effects or overdoses is that they are more likely to occur in patients prescribed the medication than those who are not.
A consistent approach is required even for searching for standardized terms. The best starting point is to get a complete ICD list in an electronic, searchable format. We create a “match list” by using principles employed in free-text searching and taking advantage of the hierarchical structure of the coding system. In addition to convention terms (like diabetes), we search for root syllables (“*diab*”), abbreviations (“DM”), synonyms, and acronyms (“AODM”) in the diagnosis and description fields. The asterisk in the search for root syllables represents a “wild card”, i.e., any combination of characters preceding or following the text of interest. We then edit the “hits” to include only the relevant terms and resort by ICD-9 code. Finally, we examine the neighboring entries to the “hits” to find related entries that might otherwise be missed. Our list of diabetes ICD-9 codes was generated using this strategy. It is stored as a table in a relational data base so that a comprehensive search can be done simply by joining a source file (e.g., problem list) to the reference table by ICD-9 code.
Searching laboratory files for diagnostic test results must also be done carefully. Standard test names are often not used, assays may not be standardized; results are expressed as text to accommodate comments (“cancelled”); many results are outside a reportable range; assays and reference values change with time and vary across sites; and there can be disagreement over what constitutes a “positive” test. We focus on abnormal HbA1c because the conditions under which glucose values are drawn are often unknown. For example, even nondiabetic patients can develop glucose 11intolerance when given large volumes of dextrose-containing intravenous fluids or re-fed after prolonged starvation. The first step is to tabulate the test names and then search for characteristic patterns (“*a1c*”), root syllables (“*hem*” for hemoglobin and “*gly*” for glyco-, glycated, or glycosylated), and abbreviations. Records containing the relevant test names are then pulled from the source file. After replacing the comments, we search for the leading characters “<“ or “>” in the results field, replace them with a null string, and then convert the text entries to numeric. Failure to follow this procedure will result in a null entry for results outside of the reportable range and a loss of patients with the most abnormal results. If results from different assays have to be compared, it is necessary to express the raw values as a percent of the upper limit of normal. The results can then be reparameterized to a common scale by multiplying the percent by the upper limit for the assay currently used. Fortunately, both the American Diabetes Association and the World Health Organization have settled upon the same value of HbA1c (≥ 6.5%) as diagnostic of diabetes.
The same issues must be addressed while searching for diabetes drugs in prescription records. Preparations may be described by their generic name, brand name, or an abbreviation; formats for tablet strength may vary (“2 mg” or “2_mg” or “2.0 mg”); the dosage form (“tablet” or “capsule”) may or may not be included; and the formulations may differ (“vial” versus “prefilled”) for the same medication. Each variation in terminology creates a separate entry even for preparations that are pharmacologically identical. The search is greatly facilitated if each prescription has been assigned to a drug class or can be linked to a standard entry in a drug reference. For example, Veterans Affairs (VA) uses the drug class HS500 for exenitide, HS501 for insulins, HS502 for oral hypoglycemic agents (OHA), and HS509 for pramlintide. Searching for these four classes achieves the same results as searching for dozens of diabetes medications.
Data Validity
The validity of data can vary widely depending upon the source. Discharge diagnoses are usually accurate because they are made by clinicians who have taken care of the patient over days if not weeks. At the NMVAHCS, only PCPs may enter problems onto the problem list, presumably after careful review of the evidence and a determination that the condition is long term and of sufficient priority to be tracked from visit to visit. Quality control is used to assure that prescription and laboratory data are accurate. On the other hand, encounter diagnoses may be entered by administrative personnel; data entry may not be standardized; results are rarely audited; diagnoses may be provisional; or they may be made after an encounter lasting only few minutes or by a clinician unfamiliar with the case. Nevertheless, external review organizations often use a small sample of outpatient encounters to assess the quality of care. By using all data sources and 100% sampling, registries may provide a much more valid assessment of an institution's processes and outcomes.
The most sophisticated electronic medical records also contain features that minimize entry errors such as logic checking, range checking, validation rules, error messages, and input masks. Logic checking compares data in two or more fields to determine if the results are compatible. For example, a high value for prostate specific antigen should not be found in a woman. Range checking consists of comparing the result to a range of plausible values. For example, an HbA1c of 20% is not possible if the reportable range for the assay is 4–18.7%. Validation rules often determine whether an entry is acceptable depending upon whether a Boolean statement evaluates to true or false. In each case, the error may be relayed to the user through error messages. An input mask presents the desired format to the user so that there is never any ambiguity about what should be entered. They are commonly used for social security numbers (XXX-XX-XXXX) or telephone numbers [(XXX) XXX-XXXX]. The user is only responsible for entering the “X's” and not the formatting characters. There is usually an option to store the entire sequence or only the characters entered. Registries are more likely to be valid if there are mechanisms to reduce entry errors at the clinical interface. Surprisingly, even advanced applications like VA's computerized patient record system (CPRS) do not have these functionalities.
Search Period
At NMVAHCS, we search for cases as far back as the system will permit. There are several reasons for this approach. Patients who have successfully managed their diabetes with diet and exercise will not have a high HbA1c, be on medications, have an outpatient encounter where diabetes is addressed, or be hospitalized for a diabetes-related condition. Moreover, this situation may persist for years. A search of recent records may miss these patients even though they are most likely to benefit from intensified preventive care. Table 2 illustrates the effect of search period on the proportion of 13,999 diabetic patients identified by the NMVAHCS registry in 2011.
TABLE 2   Impact of Search Period on Case Identification (n = 13,999)
Start date
Cases identified
FY 1997*
13,978 (99.8%)
FY 1999
13,872 (99.1%)
FY 2001
13,661 (97.6%)
FY 2003
13,408 (95.8%)
FY 2005
13,057 (93.3%)
FY 2007
12,548 (89.6%)
FY 2009
11,308 (80.8%)
FY, fiscal year.
*Fiscal year October 1996 to September 1997.
TABLE 3   Diagnosis Cohort and Current HbA1c (n = 13,999)
Cases (%)
HbA1c (mean ± SD)
966 (6.9%)
7.77 ± 1.49%
924 (6.6%)
7.41 ± 1.51%
1,089 (7.8%)
7.26 ± 1.51%
1,677 (12.0%)
7.32 ± 1.46%
1,710 (12.2%)
7.18 ± 1.51%
1,984 (14.2%)
6.87 ± 1.29%
2,613 (18.7%)
6.66 ± 1.20%
3,036 (21.7%)
6.69 ± 1.15%
Note that a search period of 8 years is required to identify greater than 95% of current patients previously diagnosed with diabetes.
Another reason for an extended search is to identify the date of diagnosis and assign the patient to an “inception cohort”. The VA Diabetes Trial has shown that diabetes of more than 20 years duration is an independent risk factor for cardiovascular event.15 Moreover, because type 2 diabetes is a progressive illness, it is reasonable to anticipate a gradual increase in HbA1c over time. This phenomenon is shown in table 3. Note that patients belonging to the 1995–1996 cohort have much higher values for HbA1c than those diagnosed in the past 2 years. This finding has major implications for provider profiling. Because established providers may have had closed panels for many years, their patients will have had diabetes much longer than recently diagnosed patients assigned to new staff. Accordingly, it is unfair to set the same standards for glycemic control for providers in two groups.
Finally, a comprehensive search of old laboratory records provides information about long-term glycemic control and the subsequent risk of microvascular complications (see Section “Advanced Functionalities”).
Sampling Bias
A sampling bias is a systematic error in the description of a population that arises from the data sources chosen for the search. Discharge diagnoses are biased toward patients with advanced complications. An elevated HbA1c misses patients who are well controlled. The medication file does not contain patients managed by diet and exercise alone. Outpatient encounters identify patients who not only have active diabetes but also have the resources to seek medical treatment. At NMVAHCS, problem lists are managed by PCPs and do not capture patients referred only for tertiary care. Thus, single sources of data not only miss certain cases but also produce results that are biased toward the ends of the severity spectrum. This phenomenon is shown in table 4.
TABLE 4   Data Sources and Quality Metrics (n = 13,999)
HbA1c not done*
HbA1c (mean ± SD)**
466 (13.3%)
7.43 ± 1.59%
HbA1c ≥6.5%
781 (7.7%)
7.35 ± 1.40%
581 (6.7%)
7.40 ± 1.48%
Problem lists
862 (7.7%)
7.12 ± 1.42%
975 (9.0%)
7.15 ± 1.44%
*Within the preceding 24 months.
**Latest value for those obtaining HbA1c in the preceding 24 months.
As expected, HbA1c is highest for cases detected through hospital records and lowest for those with a problem list entry. This observation reinforces our policy of using all five conventional data sources for the NMVAHCS diabetes registry. This strategy increases the sensitivity of the search and produces the least biased results.
Criteria Table
The NMVAHCS diabetes registry contains the first qualifying discharge date, diagnosis, and ICD-9 code; the first qualifying visit date, diagnosis, and ICD-9 code; the first qualifying problem list diagnosis, ICD-9 code, and entry date; the date and value of the first HbA1c greater than or equal to 6.5%; and the prescription type and date for the first diabetes medication. Users can search on individual criterion, the total number of criteria, or specific combinations. This option can be used to confirm diagnoses based on less reliable sources. Also, this feature facilitates the identification of patients for special initiatives (e.g., an early prevention program).
Certain data elements should be incorporated into every diabetes registry including patient identifiers, contact information, and the names of the PCP, team, and practice site. Other items may be included depending upon the purpose of the registry. For example, a registry focusing on prevention, screening, and lifestyle intervention might include:
  • Body mass index (BMI)
  • Date of the last nutritional and exercise counseling
  • Vaccination dates for influenza, pneumonia, and tetanus
  • Date and results of the last foot examination
  • Date of the last eye referral.
If the registry will be used to support intensive treatment of hyperglycemia, it might contain:
  • Most recent HbA1c
  • Prior HbA1c
  • Magnitude and rate of change
  • Most recent serum creatinine, estimated glomerular filtration rate (GFR), and urine albumin/creatinine ratio
  • Type and dose of OHA
  • Type, dose and timing of insulin injections.
If the registry will be used for macrovascular risk management, it should contain:
  • Recent tobacco history
  • Most recent blood pressure
  • Most recent low-density lipoprotein (LDL) cholesterol measurement and liver function tests
  • Current dose of aspirin, statin, and blood pressure medication.
Note that this arrangement allows the user to identify individuals who have not received appropriate testing; have not been seen for an abnormal value; or have failed to reach metabolic targets. This assessment allows the user to stratify the entire population by risk, severity, or need for services. Resources can then be managed to achieve optimal results for the entire population.
The contents of a diabetes registry should be aligned with the duties of the user. For example, if cardiovascular risk reduction is the responsibility of the diabetes team, it should contain information on LDL and lipid-lowering agents. However, if a separate team is assigned the responsibility for cardiovascular risk management for all patients, then those fields can be eliminated from the diabetes registry.
The primary purpose of diabetes registries is to improve patient outcomes. However, they are more likely to be adopted if there are benefits for clinicians as well. These benefits include identifying lapses in care and reducing malpractice exposure, reducing workload, saving time, off-loading tasks to other members of the team, providing information otherwise difficult to retrieve, triaging cases to specialty and support services, and performing complex but highly relevant calculations that clinicians are unable to do. Examples include identifying serious drug interactions that otherwise would have gone unnoticed, registry-based vaccination programs, automated referral of OHA dose titrations to the advanced practice nurse, deriving the rate of change in estimated GFR, referring cases who have repeatedly failed primary care management for a psychosocial evaluation, and estimating long-term glycemic burden from routine laboratory tests.
Poorly designed registries could significantly increase provider workload and frustration while providing minimal benefits. One example is displaying a provider's abnormal HbA1c results in a “dashboard” without determining whether they are “actionable”. “Actionable” means that the notification will ordinarily result in some remedial action on the part of the recipient. There are many reasons why an elevated HbA1c might not be actionable:
  • It was drawn from a patient for whom treatment intensification is inappropriate (e.g., terminal cancer)
  • The result represents a significant decline from a previous value indicating that the patient is already on effective treatment
  • The abnormality has already been treated
  • The abnormality has already been seen by the PCP
  • The HbA1c was drawn too soon after the last treatment change to assess the patient's response; recall that it takes several weeks for HbA1c to equilibrate
  • The same result has been observed for years suggesting that the case is a refractory treatment failure.
We used the 2011 diabetes registry of 13,999 cases at the NMVAHCS to assess the extent to which HbA1c's are “actionable” in a primary care population. The analysis was done for three cut points: ≥9.0%, ≥8.0%, and ≥7.0%. In table 5, note that only 20% of values greater than or equal to 9% were drawn at a time appropriate to evaluate a previous action. This finding suggests that those ordering the tests do not consistently review records to determine whether the results will be interpretable. In fact, HbA1c's are drawn on admission to certain units at our medical center as a matter of policy. Reviewing charts for results that are actionable would require a tremendous effort by our primary care staff. To detect the 214 actionable values ≥9, our providers would have had to review over 1,000 charts. While the yield is better for HbA1c greater than or equal to 8% or 7%, we estimate that our providers would have to review 3–4 charts to find one value that warrants intervention. Note that this requirement penalizes practices that are highly efficient in following up their laboratory tests because the number of untreated abnormalities would be small. Thus, diabetes registries should do more than retrieve data—they should process them. Advanced functionalities can be assigned to three areas:
  1. Automated triage or prioritization of cases: Triage is defined as the process of routing patients to the most appropriate level of treatment. Prioritization refers to setting the order in which cases are handled. Health systems can markedly improve their operational efficiency through registries that automatically screen, triage, and prioritize cases. These programs can be installed on the institutional servers, interrogate the data streaming into repositories from electronic medical records, and be set to update at a frequency determined by the user. The ultimate objective is to identify the health system's highest priorities at any given time and to refer those cases to the appropriate level of care.
    TABLE 5   Proportion of HbA1c That is Actionable
    Total patients
    ≥70 days after last medication change
    214 (20.4%)
    510 (24.0%)
    1,429 (31.3%)
    For example, registries can generate daily task lists for each member of a diabetes team. The administrative assistant may get a list for patients due for HbA1c testing or needing an appointment. The advanced practice nurse may get another for patients needing uptitration of an OHA, while the physician may get a third for patients needing conversion to insulin. The certified diabetes educator (CDE) may get yet another for patients failing multi-injection insulin treatment to be sure that they know how to use carbohydrate counting to adjust their rapidly-acting preparations.
    Registries are most effective when the application replicates the decisions that a clinician would make for an individual patient except that the program would perform the designated task for every patient every day. The first step is to develop a consensus among the clinicians about how a specific problem would be solved. Not only does this approach foster a spirit of collaboration between clinicians and informaticists, but it also increases the likelihood that the application would be used. The clinical process is then reduced to a decision algorithm (Figure 1) characterized by a branching sequence of decision notes. Because each decision is based on a patient attribute, the algorithm defines the data elements that should be incorporated into the registry. It then becomes a simple matter to separate the entire population into groups each of which requires a specific intervention. Each path through the algorithm represents a specific pattern of data. Queries can be written to identify patients with each specific pattern and consequently the task (if any) that should be accomplished. This approach can eliminate a large number of unnecessary chart reviews while preserving the ability to find “missed cases”. The accompanying diagram illustrates the decision algorithm used to triage elevated HbA1c's at NMVAHCS. Note that the decision notes are the boxes containing questions, while the required tasks are shown in the boxes on the right. Thus, the application prioritizes the entire diabetes population into those needing an immediate adjustment in treatment, those for whom additional testing is needed, and those for whom no action is necessary.
    zoom view
    FIGURE 1: An algorithm for triaging elevated HbA1c's.
    Although the computer program retrieves the appropriate data from regional servers, some of the questions cannot be answered by the raw data. In fact, the application uses sophisticated methodology to derive the answers for some of the nodes. Date mapping refers to placing events along a timeline to determine their sequence and intervals. At our facility, the last HbA1c is actionable if it is greater than 10 weeks after the last change in medications but not if it antedates the last change or the interval is less than or equal to 10 weeks. Registries can also be programmed to do trend analysis. For example, no action may be appropriate if the HbA1c is decreasing rapidly. The rate can be determined by “self-joining” each value to the preceding one, calculating the difference between the two readings, determining the testing interval, and dividing the former by the latter. Clinicians often find trends in data to be far more useful than the absolute values.
    Our system triages cases to members of the diabetes team using an algorithm based upon the natural history of type 2 diabetes. Most patients are initially treated with diet and exercise. When lifestyle intervention fails, OHAs are started and gradually uptitrated. When that fails, patients are started on basal insulin. Eventually, patients wind up on complicated multi-injection regimens the most sophisticated of which require carbohydrate counting. Starting a new OHA requires an assessment of indications, drug interactions, side-effects, patient acceptance, and health literacy as well as a considerable amount of education and emotional support. Because this task is complex, our software assigns this task to the PCP. On the other hand, uptitrating an existing OHA requires only an evaluation of its effectiveness and tolerance—a task assigned to the advanced practice nurse. Pharmacists might be assigned to patients on combination insulin regimens focusing on basal (e.g., premeal and bedtime) hyperglycemia. However, CDEs are assigned cases treated with lispro, glulisine, or aspart. The reason is that CDEs are the only team members at our facility skilled in carbohydrate counting used to adjust doses before meals. This triage to team members is shown in figure 2.
    zoom view
    FIGURE 2: An algorithm for triaging patients to members of a diabetes team.
    Note that, as long as the PCP approves the protocol and writes standing orders for his/her cases, the system can triage cases automatically to the most appropriate team member. This approach avoids the “bottleneck” created by having the PCP evaluate every HbA1c on every patient. The system is also designed to make sure that all team members are operating at the highest level of their abilities.
  2. Advanced clinical parameters: Computer programs can perform cutting-edge analyses of large volumes of clinical data going back many years. These analyses result in highly relevant patient assessments that have not previously been possible. For example, our diabetes registry can be linked to a chronic kidney disease registry that tracks temporal changes in estimated GFR for the entire NMVAHCS population. It also estimates the aggregate dose of nonsteroidal anti-inflammatory drugs (NSAIDs) prescribed by NMVAHCS providers over the past 15 years. The amount dispensed through each prescription is converted to an “ibuprofen equivalent” through propriety methodology, and the total is then derived from all prescriptions in the drug class. High aggregate NSAID doses have been implicated in the pathogenesis of chronic kidney and cardiovascular disease. Our diabetes registry can also be joined to a pain/narcotics registry at NMVAHCS that tracks pain scores and narcotic use. One useful feature of the latter registry is its ability to detect aberrant behaviors such as repeated early refills and prescriptions from multiple providers. The application keeps a running tabulation of the proportion of narcotics dispensed through early refills so that there is never any ambiguity about whether patients are taking their medications as prescribed. This information is highly useful to monitor narcotic use by patients with painful neuropathy.
    Perhaps the most useful of the advanced clinical parameters is an estimation of long-term glycemic burden. Basic research has shown that protein glycation (and, therefore, microvascular injury) is a function of glucose level and exposure time. It would be very useful to have a measure of long-term glycemic burden, but there is no laboratory test that spans the many years over which such injury occurs. Our application solves this problem by integrating the area under the HbA1c versus time curve using all available values in the patient's medical record. There are several uses for this information. Patients with a large glycemic burden should be assessed for subclinical microvascular injury even if they are not yet symptomatic. Unsuccessful treatment of hyperglycemia over years suggests that the primary care approach has repeatedly failed. These patients may benefit from a detailed analysis of the psychological, behavioral, or social barriers to treatment intensification before such treatment is again attempted. Finally, this feature of our software allows managers to determine if practice variation in glycemic control is correlated with microvascular complications over periods much longer than are feasible in clinical trials. In fact, this type of analysis should be the driving force for QI activities because it uncovers the relationship of local variations with local 20outcomes. Quality improvement activities that rely on published guidelines may fail because the trials that generated the “evidence” may focus on procedures or populations not relevant at the local level.
  3. Rigorous statistical treatment of the data: Computerized disease registries can perform meaningful statistical analyses of the data not previously possible. One example is to derive time-weighted averages of clinical parameters to assess clinical severity. Time-weighting is an important strategy to avoid serious biases in summary statistics that arise from the fact that measurements are taken more frequently when they are abnormal. For example, suppose that a patient is hospitalized for hyperglycemia. On the first day, six measurements of 1,000 mg/dL are obtained while treatment is initiated. On the second day, when treatment has taken effect, a single value of 100 mg/dL is obtained before the patient is discharged. The unweighted average over the seven readings for two days is [(6 × 1000) + 100]/ 7 = 871 mg/dL. However, a more reasonable estimate would be 550 mg/dL because each of the 1,000 mg/dL readings on the first day represented only 4 hours, while the single reading on the second represented 24. The correct approach is to time-weight each reading by the amount of time that it represents. Another statistical procedure is to perform transformations of skewed data so that z-scores can be interpreted as markers for “outliers”. For example, our application can be linked to a pharmacy registry that identifies drug utilization and costs for all preparations, patients, and providers in the region. The distribution of patient costs is positively-skewed, i.e., there is a long tail toward the right that drags the mean away from the median toward higher values. Calculating a standard deviation in cost for each patient is not useful because the resulting “z-score” cannot be interpreted with respect to percentile ranking. The pharmacy registry solves this problem by normalizing the cost distribution through a logarithmic transformation of raw values before calculating z-scores. Thus, a person with costs two standard deviations above the transformed mean is within the highest 2.5% of the population as a whole.
We have discussed the importance of clinician participation in the design of diabetes registries. Getting a much larger group of health care providers to use the application for routine care represents a formidable challenge. As a group, physicians have resisted the change from paper to electronic records despite the evidence that the latter improve patient outcomes. Moreover, some practices may be too stressed and overburdened to allow the implementation of novel approaches even if they improve workflow and efficiency. Although health policy and financial incentives may require the adoption of information technology, the benefits of diabetes registries will be realized only if they are embraced by the community of practitioners. How, when, or if this transition occurs is beyond the scope of this chapter.
The same aversion for information technology may make it difficult to deliver useful information to clinicians. The options include paper reports, spreadsheets (like Microsoft Excel) familiar to many casual users, desktop relational database applications (such as Microsoft Access), the reporting services of mainframe programs (like Microsoft SQL Server) or customized applications on the internet that are integrated with diabetes registries residing on institutional servers. We use the Microsoft products as examples because the company is the principal software vendor for the Department of Veterans Affairs, but many other options are available. In general, there is a trade-off between user friendliness and functionality. Early in the transition, it is reasonable to place the emphasis on the former and gradually increase the functionality of the products as clinicians become technically more sophisticated. defines beta-testing as a “second level, external pilot test of a product (usually a software) before ‘commercial quantity production’. At the beta-test stage, the product has already passed through internal validation (alpha testing) and glaring defects have been removed. But since the product may still have some minor problems that require user input, it is released to selected customers for testing under everyday conditions to spot the remaining flaws”. Another reason is to refine its aesthetic features or improve its functionalities. However, at the beta-test phase, extensive reprogramming of the application is usually not feasible. It is reasonable to perform beta-testing on a diabetes registry before it is released for use under ordinary circumstances.
Adoption will be enhanced if there are validation studies that not only demonstrate that the registry delivers accurate information but that its use also improves the efficiency or outcomes of care. Comparing the effectiveness of two care-delivery systems is the objective of comparative effectiveness research—an area that has recently been given the highest priority for research sponsored by federal agencies. Clinical trials should be done comparing registry-driven versus conventional medical care for improving prevention, treating hyperglycemia, and managing cardiovascular risk in diabetic patients. Because this intervention represents a change to the health system infrastructure, trials randomized at the patient level are probably not feasible. However, group randomization (e.g., by practice site) may be used to balance covariates at higher levels, particularly when coupled with matching or stratification for the major confounders. Differences in outcomes between registry-based and usual care can then be statistically adjusted for differences in patient attributes. It should also be remembered that patient-centric outcomes are extremely important for a chronic, potentially-disabling disease like diabetes. These outcomes include psychological, physical, or social functioning; quality of life; preservation of the activities of daily living; satisfaction with care; more informed decision making; perception of and satisfaction with health; and improved patient-provider communication. Diabetes registries have the potential for improving these patient-oriented as well as diseasespecific outcomes. Again, the federal government has placed renewed emphasis 22on this type of research through the Patient Centered Outcomes Research Institute (PCORI).
However, the success of diabetes registries will ultimately be achieved when they alert a clinician to serious problems that otherwise would be missed, increase the practice's efficiency, or improve the practitioner's outcomes, revenue, or satisfaction. At that point, the benefits change from theoretical and general to actual and personal. A diabetes registry will then become an integral part of a sustainable business model where the additional burdens of information technology are more than offset by tangible gains. This goal may be achieved only by keeping the ultimate “clients” in mind—the individual patient and his/her practitioner.
Practice Profiling
Because the registry contains all diabetic patients and their current assignments to clinicians, group practices, or sites, it is a simple matter to calculate “panel-wide” statistics for the processes and outcomes of treatment, determine average values across the institution, and identify those whose performance is significantly above or below the norm. Moreover, it is possible to stratify this analysis by or adjust for differences in patient variables (e.g., diabetes duration), physiology (e.g., BMI), or treatment (OHA versus insulin) to make the comparisons meaningful. This approach is far superior to auditing a small number of charts per provider and applying external (and perhaps irrelevant) standards without adjusting for panel characteristics.
Continuing Medical Education
Although providers choose their CME activities on the basis of interest, the new paradigm may require that providers take courses based upon their performance. Registries can be used to identify patients for case presentations, providers whose performance is suboptimal, or topics for general education. Thus, registries have the potential for aligning educational requirements with the needs of providers so that there is collective improvement in the health status of the entire population.
Meaningful Quality Improvement
Scientific evidence is only the starting point for determining how diabetic patients should be managed. Basic research and clinical trials are done under contrived conditions where psychological, behavioral, social, and financial barriers to treatment are minimized. On the other hand, practitioners must deal with these practical issues every day and often develop solutions on their own. Effective QI requires identifying practices with optimal outcomes, ascertaining the reasons, replicating successful 23solutions across the institution, and re-evaluating performance. Registries can play a central role in this iterative process.
Practice Support
We have already discussed how registries can be used to manage preventive services. Visit management consists of preparing a patient so that the encounter is not wasted for lack of data. The last testing dates for HbA1c and LDL in the registry can be reviewed to determine if they are current. If not, patients can be instructed to obtain them prior to their next visits so that the time can be spent making therapeutic decisions rather than ordering labs. Pharmacy management consists of a systematic review of drug use by the population so that pre-emptive actions can be taken. The most useful aspects of such management relate to drug safety or interactions and patient adherence. For example, the registry can be queried for patients with progressive renal insufficiency so that they can be switched from agents cleared by renal mechanisms (e.g., glyburide) to those that are not (glipizide). Medication adherence can be ascertained from prescription records. The medical possession ratio (MPR) is defined as the proportion of days for which a patient has enough medication to take the prescribed dose. It is derived by dividing the number of days supplied by the interval between the last and prior fill dates. Registries can be designed so that the MPR can be calculated for all diabetic drugs for the entire population. Counseling should then be directed at those with low levels of adherence.
An ancillary benefit of a diabetes registry is that it can serve as a sampling frame for clinical and basic research. It can be designed to evaluate the entire population for a study's inclusion and exclusion criteria so that every patient can undergo a rigorous screening process with minimal effort. A search of computerized records usually requires a waiver of informed consent and Health Insurance Portability and Accountability Act (HIPAA) authorization. A waiver is usually justified if it creates an unbiased sampling frame to maximize validity, assures equal representation of women and minorities, and is associated with minimal risk because of rigorous security precautions. At the NMVAHCS, an “honest broker” with no vested interest in the research is responsible for searching files, obtaining the approval of PCPs, and sending solicitation letters to patients. The broker is “honest” because he/she receives no benefit from the study and therefore can represent the interests of patients and the institution without a conflict of interest. Basic research can be supported by creating a sampling frame across a phenotypic spectrum. For example, the registry can be examined for “super-responders” and “nonresponders” in LDL to a given dose of a statin. Proteomics or genomics research can be done to identify the mechanisms for this phenotypic variation.
Most major industries—retailing, manufacturing, transportation, financial, hospitality, energy and natural resources, etc.—rely on information technology to perform a systematic evaluation of operations, identify where processes are suboptimal, and triage those problems to appropriate personnel for corrective action. No one would board an airplane if the pilot not only had to fly the plane but chart the course on paper using a sextant and compass. Moreover, the chances of recovery would be small if the first indication of trouble is a sudden loss in altitude. However, the US health care industry has been operating in this mode for decades—without a clear indication of what works and what does not, which patients need care and which do not, and which providers are doing well and which are not. Moreover, the first indication of trouble is often an adverse patient outcome for which there may be no recovery, exhaustion of resources resulting in withdrawal of care, or poor job satisfaction leading to a decline in the physician workforce. Fortunately, health care systems are now adopting information systems which will bring their enterprises into the modern age. Because they focus on the management of entire populations, registries are the most promising of many tools that will drive much-needed and much-delayed health care reform. As one of the most complicated chronic diseases, diabetes should be the focus of one of the first registries at almost every institution.
  1. Larsson S, Lawyer P, Garellick G, Lindahl B, Lundström M. Use of 13 disease registries in 5 countries demonstrates the potential to use outcome data to improve health care's value. Health Aff. 2012;31(1):220–7.
  1. Young RJ, Khong CK, Vaughan NJ, New J, Roxburgh M. The evolution of diabetes information systems. Diabet Med. 2002;19(suppl 4):6–12.
  1. Zai A, Grant R, Andrews C, Yee R, Chueh H. Improving diabetes population management efficiency with an informatics solution. AMIA Annu Symp Proc. 2007:1168.
  1. Khan L, Mincemoyer S, Gabbay RA. Diabetes registries: where we are and where are we headed? Diabetes Technol Ther. 2009;11(4):255–62.
  1. Russell KG, Rosenzweig J. Improving outcomes for patients with diabetes using Joslin Diabetes Center's Registry and risk stratification system. J Healthc Inf Manag. 2007;21(2):26–33.
  1. Pollard C, Bailey KA, Petitte T, Baus A, Swim M, Hendryx M. Electronic patient registries improve diabetes care and clinical outcomes in rural community health centers. J Rural Health. 2009;25(1):77–84.
  1. Coppell KJ, Anderson K, Williams SM, Lamb C, Farmer VL, Mann JI. The quality of diabetes care: a comparison between patients enrolled and not enrolled to a regional diabetes register. Prim Care Diabetes. 2011;5(2):131–7.
  1. Fischer HH, Eisert SL, Durfee MJ, Moore SL, Steele AW, McCullen K, et al. The impact of tailored diabetes registry report cards on measures of disease control: a nested randomized trial. BMC Med Inform Decis Mak. 2011;11:12.25
  1. Ciemins EL, Coon PJ, Fowles JB, Min SJ. Beyond health information technology: critical factors necessary for effective diabetes disease management. J Diabetes Sci Technol. 2009;3(3):452–60.
  1. Harris MF, Priddin D, Ruscoe W, Infante FA, O'Toole BI. Quality of care provided by general practitioners using or not using division-based diabetes registers. Med J Aust. 2002;177(5):250–2.
  1. Littenberg B, MacLean CD. Mandated diabetes registries will benefit persons with diabetes. Arch Intern Med. 2008;168(8):797–9.
  1. Trief PM, Ellison RA. Mandated diabetes registries will not benefit persons with diabetes. Arch Intern Med. 2008;168(8):799–802.
  1. Gabbay RA, Khan L, Peterson KL. Critical features for a successful implementation of a diabetes registry. Diabetes Technol Ther. 2005;7(6):958–67.
  1. Campion FX, Tully GL, Barrett JA, Andre P, Sweeney A. Improving quality of care using a diabetes registry and disease management services in an integrated delivery network. Dis Manag. 2005;8(4):245–52.
  1. Duckworth W, Abraira C, Moritz T, Reda D, Emanuele N, Reaven PD, et al. Glucose control and vascular complications in Veterans with type 2 diabetes. N Engl J Med. 2009;360:129–39.26
ICD-9 codes for diabetes at the New Mexico Veterans Affairs Health Care System (NMVAHCS)
Secondary diabetes mellitus without mention of complication, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus without mention of complication, uncontrolled
Secondary diabetes mellitus with ketoacidosis, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with ketoacidosis, uncontrolled
Secondary diabetes mellitus with hyperosmolarity, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with hyperosmolarity, uncontrolled
Secondary diabetes mellitus with other coma, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with other coma, uncontrolled
Secondary diabetes mellitus with renal manifestations, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with renal manifestations, uncontrolled
Secondary diabetes mellitus with ophthalmic manifestations, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with ophthalmic manifestations, uncontrolled
Secondary diabetes mellitus with neurological manifestations, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with neurological manifestations, uncontrolled
Secondary diabetes mellitus with peripheral circulatory disorders, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with peripheral circulatory disorders, uncontrolled
Secondary diabetes mellitus with other specified manifestations, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with other specified manifestations, uncontrolled
Secondary diabetes mellitus with unspecified complication, not stated as uncontrolled, or unspecified
Secondary diabetes mellitus with unspecified complication, uncontrolled
Diabetes mellitus without mention of complication, type II or unspecified type, not stated as uncontrolled
Diabetes mellitus without mention of complication, type I [juvenile type], not stated as uncontrolled
Diabetes mellitus without mention of complication, type II or unspecified type, uncontrolled
Diabetes mellitus without mention of complication, type I [juvenile type], uncontrolled
Diabetes with ketoacidosis, type II or unspecified type, not stated as uncontrolled
Diabetes with ketoacidosis, type I [juvenile type], not stated as uncontrolled
Diabetes with ketoacidosis, type II or unspecified type, uncontrolled
Diabetes with ketoacidosis, type I [juvenile type], uncontrolled
Diabetes with hyperosmolarity, type II or unspecified type, not stated as uncontrolled
Diabetes with hyperosmolarity, type I [juvenile type], not stated as uncontrolled
Diabetes with hyperosmolarity, type II or unspecified type, uncontrolled
Diabetes with hyperosmolarity, type I [juvenile type], uncontrolled
Diabetes with other coma, type II or unspecified type, not stated as uncontrolled
Diabetes with other coma, type I [juvenile type], not stated as uncontrolled
Diabetes with other coma, type II or unspecified type, uncontrolled
Diabetes with other coma, type I [juvenile type], uncontrolled
Diabetes with renal manifestations, type II or unspecified type, not stated as uncontrolled
Diabetes with renal manifestations, type I [juvenile type], not stated as uncontrolled
Diabetes with renal manifestations, type II or unspecified type, uncontrolled
Diabetes with renal manifestations, type I [juvenile type], uncontrolled
Diabetes with ophthalmic manifestations, type II or unspecified type, not stated as uncontrolled
Diabetes with ophthalmic manifestations, type I [juvenile type], not stated as uncontrolled
Diabetes with ophthalmic manifestations, type II or unspecified type, uncontrolled
Diabetes with ophthalmic manifestations, type I [juvenile type], uncontrolled
Diabetes with neurological manifestations, type II or unspecified type, not stated as uncontrolled
Diabetes with neurological manifestations, type I [juvenile type], not stated as uncontrolled
Diabetes with neurological manifestations, type II or unspecified type, uncontrolled
Diabetes with neurological manifestations, type I [juvenile type], uncontrolled
Diabetes with peripheral circulatory disorders, type II or unspecified type, not stated as uncontrolled
Diabetes with peripheral circulatory disorders, type I [juvenile type], not stated as uncontrolled
Diabetes with peripheral circulatory disorders, type II or unspecified type, uncontrolled
Diabetes with peripheral circulatory disorders, type I [juvenile type], uncontrolled
Diabetes with other specified manifestations, type II or unspecified type, not stated as uncontrolled
Diabetes with other specified manifestations, type I [juvenile type], not stated as uncontrolled
Diabetes with other specified manifestations, type II or unspecified type, uncontrolled
Diabetes with other specified manifestations, type I [juvenile type], uncontrolled
Diabetes with unspecified complication, type II or unspecified type, not stated as uncontrolled
Diabetes with unspecified complication, type I [juvenile type], not stated as uncontrolled
Diabetes with unspecified complication, type II or unspecified type, uncontrolled
Diabetes with unspecified complication, type II [juvenile type], uncontrolled
Polyneuropathy in diabetes
Background diabetic retinopathy
Proliferative diabetic retinopathy
Nonproliferative diabetic retinopathy NOS
Mild nonproliferative diabetic retinopathy
Moderate nonproliferative diabetic retinopathy
Severe nonproliferative diabetic retinopathy
Diabetic macular edema
Background retinopathy, unspecified
Retinal microaneurysms NOS
Retinal neovascularization NOS
Other nondiabetic proliferative retinopathy
Rubeosis iridis
Diabetic cataract
Diabetes mellitus of mother, complicating pregnancy, childbirth, or the puerperium, unspecified as to episode of care or not applicable
Diabetes mellitus of mother, with delivery
Diabetes mellitus of mother, with delivery, with mention of postpartum complication
Antepartum diabetes mellitus
Postpartum diabetes mellitus
Abnormal glucose tolerance of mother, complicating pregnancy, childbirth, or the puerperium, unspecified as to episode of care or not applicable
Abnormal glucose tolerance of mother, with delivery
Abnormal glucose tolerance of mother, with delivery, with mention of postpartum complication
Abnormal glucose tolerance of mother, antepartum
Abnormal glucose tolerance of mother, postpartum
Syndrome of infant of a diabetic mother
Neonatal diabetes mellitus
Abnormal glucose tolerance test
Impaired fasting glucose
Impaired glucose tolerance test (oral)
Other abnormal glucose
Poisoning by insulins and antidiabetic agents
Insulins and antidiabetic agents causing adverse effects in therapeutic use