Eliminating Algorithmic Racial Bias in Clinical Decision Support Algorithms: Use Cases from the Veterans Health Administration.

The Veterans Health Administration uses equity- and evidence-based principles to examine, correct, and eliminate use of potentially biased clinical equations and predictive models. We discuss the processes, successes, challenges, and next steps in four examples. We detail elimination of the race modifier for estimated kidney function and discuss steps to achieve more equitable pulmonary function testing measurement. We detail the use of equity lenses in two predictive clinical modeling tools: Stratification Tool for Opioid Risk Mitigation (STORM) and Care Assessment Need (CAN) predictive models. We conclude with consideration of ways to advance racial health equity in clinical decision support algorithms.

Association of Serious Hypoglycemic Events in Older Adults With Changes in Glycemic Performance Measures.

BACKGROUND: Reducing serious hypoglycemic events is a Federal-wide objective. Despite studies of trends for rates of serious hypoglycemia in existing literature, rigorous evaluation of links between the observed trends and changes in professional guidelines or performance measures for glycemic control is lacking. OBJECTIVE: To evaluate whether changes in professional society guidelines and performance measures for glycemic control correspond to changes in rates of serous hypoglycemia. RESEARCH DESIGN: This was a retrospective observational study. We merged Veterans Health Administration (VHA) and Medicare patient-level databases of VHA patients and identified those aged 65 years and above and receiving hypoglycemic agents. We derived age-adjusted and sex-adjusted annual rates and constructed piecewise Poisson regression models adjusting for age and sex to assess time trends of the rates. SUBJECTS: VHA patients, 2002-2015. MEASURES: The main outcome was the annual rates (2004-2015) of serious hypoglycemia, defined as hypoglycemia-related emergency department visits or hospitalizations. Secondary outcomes were annual rates of hemoglobin (Hb) A1c level <7% and >9%. Age and sex were additional variables. RESULTS: The annual rate for hypoglycemia decreased by 4.8% (rate ratio: 0.952; 95% confidence interval, 0.949-0.956) for 2008-2015 but did not change (1.001; 0.994-1.001) in 2004-2008. In 2008-2015, the annual rate for HbA1c <7% decreased by 5.0% (0.950; 0.949-0.951) but for HbA1c >9%, increased by 7.9% (1.079; 1.076-1.082). CONCLUSION: The cooccurrence of decreasing rates for HbA1c<7% and serious hypoglycemia since 2008 supports the possibility that withdrawal of a <7% HbA1c measure in 2008 impacted clinical practice and patient outcomes.

Racial differences in trends of serious hypoglycemia among higher risk older adults in US Veterans Health Administration, 2004-2015: Relationship to comorbid conditions, insulin use, and hemoglobin A1c level.

AIMS: To evaluate temporal trends in racial/ethnic groups in rates of serious hypoglycemia among higher risk patients dually enrolled in Veterans Health Administration and Medicare fee-for-service and assess the relationship(s) between hypoglycemia rates, insulin/secretagogues and comorbid conditions. METHODS: Retrospective observational serial cross-sectional design. Patients were ≥65 years receiving insulin and/or secretagogues. The primary outcome was the annual (period prevalence) rates (2004-2015), per 1000 patient years, of serious hypoglycemic events, defined as hypoglycemic-related emergency department visits or hospitalizations. RESULTS: Subjects were 77-83% White, 7-10% Black, 4-5% Hispanic, <2% women; 38-58% were ≥75 years old; 72-75% had ≥1 comorbidity. In 2004-2015, rates declined from 63.2 to 33.6(-46.9%) in Blacks; 29.7 to 20.3 (-31.6%) in Whites; and 41.8 to 29.6 (-29.3%) in Hispanics. The Black-White rate differences narrowed regardless of insulin use, hemoglobin A1c level, and frequency and various combinations of comorbid conditions. Among insulin users, the Black-White contrast decreased from 34.7 (98.5 vs. 63.8) in 2004 to 13.2 (43.6 vs. 30.4) in 2015; in non-insulin users, the contrast was 25.7 (44.1 vs. 18.4) in 2004 and 10.1 (18.9 vs. 8.8) in 2015. CONCLUSION: Marked declines in serious hypoglycemia events occurred across race, medications, and comorbidities, suggesting significant changes in clinical practice.

Balancing measures: identifying unintended consequences of diabetes quality performance measures in patients at high risk for hypoglycemia.

OBJECTIVE: To determine if changes in overtreatment rates were associated with changes in undertreatment rates. DESIGN: Pre-test/post-test study used cross-sectional administrative data from calendar years (CYs) 2013 and 2016. SETTING: The Veterans Health Administration. PARTICIPANTS: Patients with diabetes at risk for hypoglycemia (n = 171 875 and 166 703 in 2013 and 2016, respectively). INTERVENTION: Observational study of extant initiatives to reduce overtreatment. MAIN OUTCOME MEASURES: Overtreatment rate of diabetes defined at the proportion of patients in the group at high risk for hypoglycemia with A1c < 7.0%. Undertreatment defined as A1C > 9%. RESULTS: There was marked variation in overtreatment rates; for A1c < 7%, overtreatment rates ranged from 26.4% to 58.2% and 26.2% to 49.2% at the facility level in 2013 and 2016, respectively. The mean (±standard deviation (SD)) facility-level overtreatment rates fell from 40.3 (±5.3)% in 2013 to 37.75 (±4.70)% in 2016 (P < 0.001, paired t-test). Facility undertreatment rates ranged from 5.8% to 16.9% and 6.8% to 18.7% at the facility level in 2013 and 2016, respectively. The mean (±SD) undertreatment rate rose from 10.3 (±2.2)% in 2013 to 11.0 (±2.4)% in 2016 (P ≤ 0.001, paired t-test). However, change at individual facilities ranged from a decrease of 4.6% to an increase of 7.2%. Within year correlations were stronger than between year correlations. Overtreatment defined as A1c < 7 in this population inversely correlated strongly with undertreatment (r = -0.653, P < 0.001). CONCLUSIONS: Promotion of overtreatment reduction may be associated with an increase in undertreatment in patients with diabetes. Unintended consequence should be considered when implementing and evaluating quality measures and systems should include balancing measures to identify potential unintended harms.

Defining and measuring population health quality of outpatient diabetes care in Israel: lessons from the quality indicators in community health program.

In Israel, as in other Organization for Economic Co-operation and Development countries, performance measurement is a key public health strategy in monitoring and improving population health outcomes. The Israeli Quality Indicators in Community Healthcare (QICH) program has utilized electronic health records to monitor ambulatory care for the entire Israeli population since 2002. In 2006 the measures were updated to include laboratory values. They have been subsequently revised by stratifying by age, duration, adding medications, and changing frequency of testing for certain process measures. However, the QICH glycemic control measures do not address co-morbid conditions either thru exclusion criteria or higher target ranges. They also do not address potential over treatment in patients with complex medication conditions.In the United States there have also been changes in nationally endorsed diabetes specific performance measures since 2007. However, there have also been public disagreements among United States professional societies, government agencies, and performance measurement organizations as to whether the current glycemic dichotomous ("all or none") threshold measures, without exclusion criteria, are consistent with the most recent evidence. Specifically, most guidelines now recommend individualized target goals based upon co-morbid conditions, risk of harms from medications, and patient preferences.Concerns have been raised that the current glycemic performance measures have resulted in inappropriate care, such as medication over-treatment, and serious harms, such as hypoglycemia, especially in older adults. There currently are no national surveillance systems or measures that monitor these untoward outcomes.We recommend several actions that QICH could consider to advance diabetes specific performance measurement science and population health: Convene an international conference; implement technical modifications of current measures and surveillance systems; and, most importantly, acknowledge patient autonomy by developing measures that document individualization of target values using shared decision making.

Using Shared Decision-Making to Address Possible Overtreatment in Patients at High Risk for Hypoglycemia: The Veterans Health Administration's Choosing Wisely Hypoglycemia Safety Initiative.

IN BRIEF Successful management of patients with diabetes requires individualizing A1C and treatment goals in conjunction with identifying and managing hypoglycemia risk. This article describes the Veterans Health Administration's Choosing Wisely Hypoglycemia Safety Initiative (CW-HSI), a voluntary program that aims to reduce the occurrence of hypoglycemia through shared decision-making about deintensifying diabetes treatment in a dynamic cohort of patients identified as being at high risk for hypoglycemia and potentially overtreated. The CW-HSI incorporates education for patients and clinicians, as well as clinical decision support tools, and has shown decreases in the proportions of high-risk patients potentially overtreated and impacts on the frequency of reported hypoglycemia.

An expanded prevention quality diabetes composite: Quantifying the burden of preventable hospitalizations for older adults with diabetes.

AIM: To expand the existing United States Agency for Health Research and Quality (AHRQ) Diabetes composite (AHRQ-DC) to include additional preventable hospitalizations specific or relevant to diabetes. METHODS: A cross-sectional analysis of 834,696 veteran patients with diabetes aged ≥65 years in 2012. An Expanded Diabetes Composite (Expanded-DC) was developed utilizing: (1) the diabetes-specific category: the AHRQ-DC (short-term and long-term complications, uncontrolled diabetes, lower extremity amputations) and two proposed conditions: hypoglycemia and lower extremity ulcers/inflammation/infections (LEU) and (2) the diabetes-relevant category: the AHRQ-Acute Composite (dehydration, pneumonia, urinary tract infections) and one proposed condition, acute kidney injury (AKI). RESULTS: The study population was 98% male, 80% White, 10% Black, and 5% Hispanic; 71% had complex comorbidities. There were 64,243 (77.0 admissions/1000 patients) hospitalizations in the Expanded-DC, compared to 13,523 (16.2) in the AHRQ-DC, a 4.7 fold increase. Hospitalizations from AHRQ-Acute Composite and the three proposed conditions added 79% to the Expanded-DC. LEU and hypoglycemia added 39% to the diabetes-specific category. AKI added 18% to the diabetes-relevant category. Blacks incurred more preventable hospitalizations (85.9) than Whites (74.7); as did patients with complex comorbidities (93.6) versus those without (34.6). CONCLUSION: The AHRQ-DC substantially underestimates rates of clinically important preventable hospitalizations in older diabetes patients.

State-Level Variability in Veteran Reliance on Veterans Health Administration and Potentially Preventable Hospitalizations: A Geospatial Analysis.

Most Veterans who use the Veterans Health Administration (VHA) also utilize private-sector health care providers. To better inform local and regional health care planning, we assessed the association between reliance on VHA ambulatory care and total and system-specific preventable hospitalization rates (PHRs) at the state level. We conducted a retrospective dynamic cohort study using Veterans with diabetes mellitus, aged 66 years or older, and dually enrolled in VHA and Medicare parts A and B from 2004 to 2010. While controlling for median age and proportion of males, we measured the association between reliance on VHA ambulatory care and PHRs at the state level using multivariable ordinary least square regression, geographically weighted regression, and generalized additive models. We measured geospatial patterns in PHRs using global Moran's I and univariate local indicator spatial analysis. Approximately 30% of hospitalized Veterans experienced a preventable hospitalization. Reliance on VHA ambulatory care at the state level ranged from 13.92% to 67.78% and was generally not associated with PHRs. Geospatial analysis consistently identified a cluster of western states with low PHRs from 2006 to 2010. Given the generally low reliance on VHA ambulatory care and lack of association between this reliance and PHRs, policy changes to improve Veterans' health care outcomes should address private-sector care in addition to VHA care.

How the dual process model of human cognition can inform efforts to de-implement ineffective and harmful clinical practices: A preliminary model of unlearning and substitution.

RATIONALE AND OBJECTIVES: One way to understand medical overuse at the clinician level is in terms of clinical decision-making processes that are normally adaptive but become maladaptive. In psychology, dual process models of cognition propose 2 decision-making processes. Reflective cognition is a conscious process of evaluating options based on some combination of utility, risk, capabilities, and/or social influences. Automatic cognition is a largely unconscious process occurring in response to environmental or emotive cues based on previously learned, ingrained heuristics. De-implementation strategies directed at clinicians may be conceptualized as corresponding to cognition: (1) a process of unlearning based on reflective cognition and (2) a process of substitution based on automatic cognition. RESULTS: We define unlearning as a process in which clinicians consciously change their knowledge, beliefs, and intentions about an ineffective practice and alter their behaviour accordingly. Unlearning has been described as "the questioning of established knowledge, habits, beliefs and assumptions as a prerequisite to identifying inappropriate or obsolete knowledge underpinning and/or embedded in existing practices and routines." We hypothesize that as an unintended consequence of unlearning strategies clinicians may experience "reactance," ie, feel their professional prerogative is being violated and, consequently, increase their commitment to the ineffective practice. We define substitution as replacing the ineffective practice with one or more alternatives. A substitute is a specific alternative action or decision that either precludes the ineffective practice or makes it less likely to occur. Both approaches may work independently, eg, a substitute could displace an ineffective practice without changing clinicians' knowledge, and unlearning could occur even if no alternative exists. For some clinical practice, unlearning and substitution strategies may be most effectively used together. CONCLUSIONS: By taking into account the dual process model of cognition, we may be able to design de-implementation strategies matched to clinicians' decision-making processes and avoid unintended consequence.

Identification of outliers and positive deviants for healthcare improvement: looking for high performers in hypoglycemia safety in patients with diabetes.

BACKGROUND: The study objectives were to determine: (1) how statistical outliers exhibiting low rates of diabetes overtreatment performed on a reciprocal measure - rates of diabetes undertreatment; and (2) the impact of different criteria on high performing outlier status. METHODS: The design was serial cross-sectional, using yearly Veterans Health Administration (VHA) administrative data (2009-2013). Our primary outcome measure was facility rate of HbA1c overtreatment of diabetes in patients at risk for hypoglycemia. Outlier status was assessed by using two approaches: calculating a facility outlier value within year, comparator group, and A1c threshold while incorporating at risk population sizes; and examining standardized model residuals across year and A1c threshold. Facilities with outlier values in the lowest decile for all years of data using more than one threshold and comparator or with time-averaged model residuals in the lowest decile for all A1c thresholds were considered high performing outliers. RESULTS: Using outlier values, three of the 27 high performers from 2009 were also identified in 2010-2013 and considered outliers. There was only modest overlap between facilities identified as top performers based on three thresholds: A1c < 6%, A1c < 6.5%, and A1c < 7%. There was little effect of facility complexity or regional Veterans Integrated Service Networks (VISNs) on outlier identification. Consistent high performing facilities for overtreatment had higher rates of undertreatment (A1c > 9%) than VA average in the population of patients at high risk for hypoglycemia. CONCLUSIONS: Statistical identification of positive deviants for diabetes overtreatment was dependent upon the specific measures and approaches used. Moreover, because two facilities may arrive at the same results via very different pathways, it is important to consider that a "best" practice may actually reflect a separate "worst" practice.

Synopsis of the 2017 U.S. Department of Veterans Affairs/U.S. Department of Defense Clinical Practice Guideline: Management of Type 2 Diabetes Mellitus.

DESCRIPTION: In April 2017, the U.S. Department of Veterans Affairs (VA) and the U.S. Department of Defense (DoD) approved a joint clinical practice guideline for the management of type 2 diabetes mellitus. METHODS: The VA/DoD Evidence-Based Practice Work Group convened a joint VA/DoD guideline development effort that included a multidisciplinary panel of practicing clinician stakeholders and conformed to the Institute of Medicine's tenets for trustworthy clinical practice guidelines. The guideline panel developed key questions in collaboration with the ECRI Institute, which systematically searched and evaluated the literature through June 2016, developed an algorithm, and rated recommendations by using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) system. RECOMMENDATIONS: This synopsis summarizes key features of the guideline in 7 areas: patient-centered care and shared decision making, glycemic biomarkers, hemoglobin A1c target ranges, individualized treatment plans, outpatient pharmacologic treatment, glucose targets for critically ill patients, and treatment of hospitalized patients.

Provider and Site-Level Determinants of Testosterone Prescribing in the Veterans Healthcare System.

Context: Testosterone prescribing rates have increased substantially in the past decade. However, little is known about the context within which such prescriptions occur. Objective: We evaluated provider- and site-level determinants of receipt of testosterone and of guideline-concordant testosterone prescribing. Design: This study was cross-sectional in design. Setting: This study was conducted at the Veterans Health Administration (VA). Participants: Study participants were a national cohort of male patients who had received at least one outpatient prescription within the VA during fiscal year (FY) 2008 to FY 2012. A total of 38,648 providers and 130 stations were associated with these patients. Main Outcome Measure: This study measured receipt of testosterone and guideline-concordant testosterone prescribing. Results: Providers ranging in age from 31 to 60 years, with less experience in the VA [all adjusted odds ratio (AOR), <2; P < 0.01] and credentialed as medical doctors in endocrinology (AOR, 3.88; P < 0.01) and urology (AOR, 1.48; P < 0.01) were more likely to prescribe testosterone compared with older providers, providers of longer VA tenure, and primary care providers, respectively. Sites located in the West compared with the Northeast [AOR, 1.75; 95% confidence interval (CI), 1.45-2.11] and care received at a community-based outpatient clinic compared with a medical center (AOR, 1.22; 95% CI, 1.20-1.24) also predicted testosterone use. Although they were more likely to prescribe testosterone, endocrinologists were also more likely to obtain an appropriate workup before prescribing compared with primary care providers (AOR, 2.14; 95% CI, 1.54-2.97). Conclusions: Our results highlight the opportunity to intervene at both the provider and the site levels to improve testosterone prescribing. This study also provides a useful example of how to examine contributions to prescribing variation at different levels of the health care system.

A Proposal for an Out-of-Range Glycemic Population Health Safety Measure for Older Adults With Diabetes.

OBJECTIVE: To evaluate patient-level glycemic control and facility variation of a proposed out-of-range (OOR) measure (overtreatment [OT] [HbA1c <7% (53 mmol/mol)] or undertreatment [UT] [>9% (75 mmol/mol)]) compared with the standard measure (SM) (HbA1c <8% [64 mmol/mol]) in high-risk older adults. RESEARCH DESIGN AND METHODS: Veterans Health Administration patients ≥65 years of age in 2012 who were taking antihyperglycemic agents in 2013 were identified. Patient-level rates and facility-level rates/rankings were calculated by age and comorbid illness burden. RESULTS: We identified 303,097 patients who were taking antiglycemic agents other than metformin only. The study population comprised 193,689 patients with at least one significant medical, neurological, or mental health condition; 98.2% were taking a sulfonylurea and/or insulin; 55.2% were aged 65-75 years; and 44.8% were aged >75 years. The 47.4% of patients 65-75 years met the OOR measure (33.4% OT, 14% UT), and 65.7% met the SM. For patients aged >75 years, rates were 48.1% for OOR (39.2% OT; 8.9% UT) and 73.2% for SM. Facility-level rates for OOR for patients aged 65-75 years ranged from 33.7 to 60.4% (median 47.4%), with a strong inverse correlation (ρ = -0.41) between SM and OOR performance rankings. Among the best-performing 20% facilities on the SM, 14 of 28 ranked in the worst-performing 20% on the OOR measure; 12 of 27 of the worst-performing 20% facilities on the SM ranked in the best-performing 20% on the OOR measure. CONCLUSIONS: Facility rankings that are based on an SM (potential benefits) and OOR measure (potential risks) differ substantially. An OOR for high-risk populations can focus quality improvement on individual patient evaluation to reduce the risk for short-term harms.

Who Gets Testosterone? Patient Characteristics Associated with Testosterone Prescribing in the Veteran Affairs System: a Cross-Sectional Study.

BACKGROUND: There has been concern about the growing off-label use of testosterone. Understanding the context within which testosterone is prescribed may contribute to interventions to improve prescribing. OBJECTIVE: To evaluate patient characteristics associated with receipt of testosterone. DESIGN: Cross-sectional. SETTING: A national cohort of male patients, who had received at least one outpatient prescription within the Veterans Affairs (VA) system during Fiscal Year 2008- Fiscal Year 2012. PARTICIPANTS: The study sample consisted of 682,915 non-HIV male patients, of whom 132,764 had received testosterone and a random 10% sample, 550,151, had not. MAIN MEASURES: Conditions and medications associated with testosterone prescription. KEY RESULTS: Only 6.3% of men who received testosterone from the VA during the study period had a disorder of the testis, pituitary or hypothalamus associated with male hypogonadism. Among patients without a diagnosed disorder of hypogonadism, the use of opioids and obesity were the strongest predictors of testosterone prescription. Patients receiving >100 mg/equivalents of oral morphine daily (adjusted odds ratio = 5.75, p < 0.001) and those with body mass index (BMI) >40 kg/m2 (adjusted odds ratio = 3.01, p < 0.001) were more likely to receive testosterone than non-opioid users and men with BMI <25 kg/m2. Certain demographics (age 40-54, White race), comorbid conditions (sleep apnea, depression, and diabetes), and medications (antidepressants, systemic corticosteroids) also predicted a higher likelihood of testosterone receipt, all with an adjusted odds ratio less than 2 (p < 0.001). CONCLUSIONS: In the VA, 93.7% of men receiving testosterone did not have a diagnosed condition of the testes, pituitary, or hypothalamus. The strongest predictors of testosterone receipt (e.g., obesity, receipt of opioids), which though are associated with unapproved, off-label use, may be valid reasons for therapy. Interventions should aim to increase the proportion of testosterone recipients who have a valid indication.

Overview and Discussion of the 2017 VA/DoD Clinical Practice Guideline for the Management of Type 2 Diabetes Mellitus in Primary Care.

The 2017 diabetes mellitus guidelines emphasize shared decision making, dietary changes, and HbA1c target range for optimal control of diabetes mellitus.

Characteristics of High-Functioning Collaborations Between Primary Care and Podiatry in VHA Patient Aligned Care Teams.

The key to high-functioning PACT/Podiatry teams rests with the quality of the communication between providers.

Variability in estimated glomerular filtration rate values is a risk factor in chronic kidney disease progression among patients with diabetes.

BACKGROUND: It is unknown whether variability of estimated Glomerular Filtration Rate (eGFR) is a risk factor for dialysis or death in patients with chronic kidney disease (CKD). This study aimed to evaluate variability of estimated Glomerular Filtration Rate (eGFR) as a risk factor for dialysis or death to facilitate optimum care among high risk patients. METHODS: A longitudinal retrospective cohort study of 70,598 Veterans Health Administration veteran patients with diabetes and CKD (stage 3-4) in 2000 with up to 5 years of follow-up. VHA and Medicare files were linked to derive study variables. We used Cox proportional hazards models to evaluate association between time to initial dialysis/death and key independent variables: time-varying eGFR variability (measured by standard deviation (SD)) and eGFR means and slopes while adjusting for prior hospitalizations, and comorbidities. RESULTS: There were 76.7% older than 65 years, 97.5% men, and 81.9% Whites. Patients were largely in early stage 3 (61.2%), followed by late stage 3 (28.9%), and stage 4 (9.9%); 29.1%, 46.8%, and 73.3%, respectively, died or had dialysis during the follow-up. eGFR SDs (median: 5.8, 5.1, and 4.0 ml/min/1.73 m(2)) and means (median: 54.1, 41.0, 27.2 ml/min/1.73 m(2)) from all two-year moving intervals decreased as CKD advanced; eGFR variability (relative to the mean) increased when CKD progressed (median coefficient of variation: 10.9, 12.8, and 15.4). Cox regressions revealed that one unit increase in a patient's standard deviation of eGFRs from prior two years was significantly associated with about 7% increase in risk of dialysis/death in the current year, similarly in all three CKD stages. This was after adjusting for concurrent means and slopes of eGFRs, demographics, prior hospitalization, and comorbidities. For example, the hazard of dialysis/death increased by 7.2% (hazard ratio:1.072; 95% CI = 1.067, 1.080) in early stage 3. CONCLUSION: eGFR variability was independently associated with elevated risk of dialysis/death even after controlling for eGFR means and slopes.