Case Studies for Overcoming Challenges in Using Big Data in Cancer.

The analysis of big healthcare data has enormous potential as a tool for advancing oncology drug development and patient treatment, particularly in the context of precision medicine. However, there are challenges in organizing, sharing, integrating, and making these data readily accessible to the research community. This review presents five case studies illustrating various successful approaches to addressing such challenges. These efforts are CancerLinQ, the American Association for Cancer Research Project GENIE, Project Data Sphere, the National Cancer Institute Genomic Data Commons, and the Veterans Health Administration Clinical Data Initiative. Critical factors in the development of these systems include attention to the use of robust pipelines for data aggregation, common data models, data deidentification to enable multiple uses, integration of data collection into physician workflows, terminology standardization and attention to interoperability, extensive quality assurance and quality control activity, incorporation of multiple data types, and understanding how data resources can be best applied. By describing some of the emerging resources, we hope to inspire consideration of the secondary use of such data at the earliest possible step to ensure the proper sharing of data in order to generate insights that advance the understanding and the treatment of cancer.

Chlorthalidone vs. Hydrochlorothiazide for Hypertension-Cardiovascular Events.

BACKGROUND: Whether chlorthalidone is superior to hydrochlorothiazide for preventing major adverse cardiovascular events in patients with hypertension is unclear. METHODS: In a pragmatic trial, we randomly assigned adults 65 years of age or older who were patients in the Department of Veterans Affairs health system and had been receiving hydrochlorothiazide at a daily dose of 25 or 50 mg to continue therapy with hydrochlorothiazide or to switch to chlorthalidone at a daily dose of 12.5 or 25 mg. The primary outcome was a composite of nonfatal myocardial infarction, stroke, heart failure resulting in hospitalization, urgent coronary revascularization for unstable angina, and non-cancer-related death. Safety was also assessed. RESULTS: A total of 13,523 patients underwent randomization. The mean age was 72 years. At baseline, hydrochlorothiazide at a dose of 25 mg per day had been prescribed in 12,781 patients (94.5%). The mean baseline systolic blood pressure in each group was 139 mm Hg. At a median follow-up of 2.4 years, there was little difference in the occurrence of primary-outcome events between the chlorthalidone group (702 patients [10.4%]) and the hydrochlorothiazide group (675 patients [10.0%]) (hazard ratio, 1.04; 95% confidence interval, 0.94 to 1.16; P = 0.45). There were no between-group differences in the occurrence of any of the components of the primary outcome. The incidence of hypokalemia was higher in the chlorthalidone group than in the hydrochlorothiazide group (6.0% vs. 4.4%, P<0.001). CONCLUSIONS: In this large pragmatic trial of thiazide diuretics at doses commonly used in clinical practice, patients who received chlorthalidone did not have a lower occurrence of major cardiovascular outcome events or non-cancer-related deaths than patients who received hydrochlorothiazide. (Funded by the Veterans Affairs Cooperative Studies Program; ClinicalTrials.gov number, NCT02185417.).

Design of a pragmatic clinical trial embedded in the Electronic Health Record: The VA's Diuretic Comparison Project.

BACKGROUND: Recent US guidelines recommend chlorthalidone over other thiazide-type diuretics for the treatment of hypertension based on its long half-life and proven ability to reduce CVD events. Despite recommendations most clinicians prescribe hydrochlorothiazide (HCTZ) over chlorthalidone (CTD). No randomized controlled data exist comparing these two diuretics on cardiovascular outcomes. METHODS: The Diuretic Comparison Project (DCP) is a multicenter, two-arm, parallel, Prospective Randomized Open, Blinded End-point (PROBE) trial testing the primary hypothesis that CTD is superior to HCTZ in the prevention of non-fatal CVD events and non-cancer death. Patients with hypertension taking HCTZ 25 or 50 mg were randomly assigned to either continue their current HCTZ or switch to an equipotent dose of CTD. The primary outcome is time to the first occurrence of a composite outcome consisting of a non-fatal CVD event (stroke, myocardial infarction, urgent coronary revascularization because of unstable angina, or hospitalization for acute heart failure) or non-cancer death. The trial randomized 13,523 patients at 72 VA medical centers. The study is conducted by a centralized research team with site procedures embedded in the electronic health record and all data collected through administrative claims data, with no study related visits for participants. The trial will have 90% power to detect an absolute reduction in the composite event rate of 2.4%. RESULTS: Enrollment ended in November 2021. There are 4128 participting primary care providers and 16,595 patients individually consented to participate, 13,523 of whom were randomized. CONCLUSIONS: DCP should provide much needed evidence as to whether CTD is superior to HCTZ in preventing cardiovascular events in hypertensive patients. CLINICAL TRIAL REGISTRATION: NCT02185417 [https://clinicaltrials.gov/ct2/show/NCT02185417].

Identification of Acute Decompensated Heart Failure Hospitalizations Using Administrative Data.

Hospitalization for acute decompensated heart failure (ADHF) is an important outcome in clinical trials and heart failure registries; however, the optimal strategy to identify these hospitalizations using International Classification of Diseases, Ninth Revision (ICD-9) codes is uncertain. We sought to identify diagnostic codes that improve ascertainment of ADHF hospitalizations. Heart failure-related ICD-9 principal discharge codes were used to identify 2,202 hospitalizations within the Minneapolis Veterans Affairs Medical Center from 2009 to 2014. Two independent reviewers adjudicated 447 of these hospitalizations to determine the accuracy of each code. We then applied our findings to an unadjusted nationwide sample containing the same ICD-9 codes of interest, from which overall positive predictive value (PPV), sensitivity, and accuracy were calculated. Use of 428.x alone resulted in a PPV of 91.3% (95% confidence interval [CI] 91.0 to 91.7), sensitivity of 97.5% (95% CI 97.3 to 97.6), and accuracy of 89.7% (95% CI 89.4 to 90.0). Combining 428.x with 402.x1, 404.x1, 415, and 518.4 resulted in improved sensitivity (99.2%; 95% CI 99.0 to 99.3) and accuracy (90.7%; 95% CI 90.4 to 91.1) while maintaining a PPV of 91.1% (95% CI 90.7 to 91.4). Excluding chronic heart failure codes (428.22, 428.32, and 428.42) from the proposed strategy resulted in an improvement of PPV to 92.3% (95% CI 92.0 to 92.6), although sensitivity and accuracy decreased to 96.6% (95% CI 96.3 to 96.8) and 90.0% (95% CI 89.6 to 90.3), respectively. In conclusion, a combination of codes including 428.x, 402.x1, 404.x1, 415, and 518.4 improves sensitivity and overall accuracy in ascertaining ADHF events compared with 428.x alone. This strategy could be further improved by manual adjudication of chronic heart failure codes.

The VA Point-of-Care Precision Oncology Program: Balancing Access with Rapid Learning in Molecular Cancer Medicine.

The Department of Veterans Affairs (VA) recognized the need to balance patient-centered care with responsible creation of generalizable knowledge on the effectiveness of molecular medicine tools. Embracing the principles of the rapid learning health-care system, a new clinical program called the Precision Oncology Program (POP) was created in New England. The POP integrates generalized knowledge about molecular medicine in cancer with a database of observations from previously treated veterans. The program assures access to modern genomic oncology practice in the veterans affairs (VA), removes disparities of access across the VA network of clinical centers, disseminates the products of learning that are generalizable to non-VA settings, and systematically presents opportunities for patients to participate in clinical trials of targeted therapeutics.

Measuring use of evidence based psychotherapy for posttraumatic stress disorder.

To improve methods of estimating use of evidence-based psychotherapy for posttraumatic stress disorder in the Veteran's health administration, we evaluated administrative data and note text for patients newly enrolling in six VHA outpatient PTSD clinics in New England during the 2010 fiscal year (n = 1,924). Using natural language processing, we developed machine learning algorithms that mimic human raters in classifying note text. We met our targets for algorithm performance as measured by precision, recall, and F-measure. We found that 6.3 % of our study population received at least one session of evidence-based psychotherapy during the initial 6 months of treatment. Evidence-based psychotherapies appear to be infrequently utilized in VHA outpatient PTSD clinics in New England. Our method could support efforts to improve use of these treatments.

Risk of community-acquired pneumonia in veteran patients to whom proton pump inhibitors were dispensed.

BACKGROUND: Observational studies linking proton pump inhibitor (PPI) exposure with community-acquired pneumonia (CAP) have reported either modest or no associations. Accordingly, we studied PPI exposure and CAP in veteran patients, using a retrospective, nested case-control design. METHODS: From linked pharmacy and administrative databases of the New England Veterans Healthcare System, we identified 71985 outpatients newly prescribed PPIs between 1998 and 2007; 1544 patients met criteria for CAP subsequent to PPI initiation; 15440 controls were matched through risk-set sampling by age and time under observation. Crude and adjusted odds ratios comparing current with past PPI exposures, as well as tests for interactions, were conducted for the entire and stratified samples. RESULTS: Current PPI use associated with CAP (adjusted odds ratio [OR], 1.29 [95% confidence interval {CI}, 1.15-1.45]). Risks were not substantially altered by age or year of diagnosis. Dementia (n = 85; P = .062 for interaction) and sedative/tranquilizer use (n = 224; P = .049 for interaction) were likely effect modifiers increasing a PPI-CAP association; conversely, for some chronic medical conditions, PPI-associated CAP risks were reversed. PPI exposures between 1 and 15 days increased CAP risks, compared with longer exposures, but PPI initiation also frequently occurred shortly after CAP diagnoses. Prescribed PPI doses >1 dose/day also increased PPI-associated CAP risks. CONCLUSIONS: Among the veterans studied, current compared with past PPI exposures associated modestly with increased risks of CAP. However, our observations that recent treatment initiation and higher PPI doses were associated with greater risks, and the inconsistent PPI-CAP associations between patient subgroups, indicate that further inquiries are needed to separate out coincidental patterns of associations.

Automated concept-level information extraction to reduce the need for custom software and rules development.

OBJECTIVE: Despite at least 40 years of promising empirical performance, very few clinical natural language processing (NLP) or information extraction systems currently contribute to medical science or care. The authors address this gap by reducing the need for custom software and rules development with a graphical user interface-driven, highly generalizable approach to concept-level retrieval. MATERIALS AND METHODS: A 'learn by example' approach combines features derived from open-source NLP pipelines with open-source machine learning classifiers to automatically and iteratively evaluate top-performing configurations. The Fourth i2b2/VA Shared Task Challenge's concept extraction task provided the data sets and metrics used to evaluate performance. RESULTS: Top F-measure scores for each of the tasks were medical problems (0.83), treatments (0.82), and tests (0.83). Recall lagged precision in all experiments. Precision was near or above 0.90 in all tasks. Discussion With no customization for the tasks and less than 5 min of end-user time to configure and launch each experiment, the average F-measure was 0.83, one point behind the mean F-measure of the 22 entrants in the competition. Strong precision scores indicate the potential of applying the approach for more specific clinical information extraction tasks. There was not one best configuration, supporting an iterative approach to model creation. CONCLUSION: Acceptable levels of performance can be achieved using fully automated and generalizable approaches to concept-level information extraction. The described implementation and related documentation is available for download.

A point-of-care clinical trial comparing insulin administered using a sliding scale versus a weight-based regimen.

BACKGROUND: Clinical trials are widely considered the gold standard in comparative effectiveness research (CER) but the high cost and complexity of traditional trials and concerns about generalizability to broad patient populations and general clinical practice limit their appeal. Unsuccessful implementation of CER results limits the value of even the highest quality trials. Planning for a trial comparing two standard strategies of insulin administration for hospitalized patients led us to develop a new method for a clinical trial designed to be embedded directly into the clinical care setting thereby lowering the cost, increasing the pragmatic nature of the overall trial, strengthening implementation, and creating an integrated environment of research-based care. PURPOSE: We describe a novel randomized clinical trial that uses the informatics and statistics infrastructure of the Veterans Affairs Healthcare System (VA) to illustrate one key component (called the point-of-care clinical trial - POC-CT) of a 'learning healthcare system,' and settles a clinical question of interest to the VA. METHODS: This study is an open-label, randomized trial comparing sliding scale regular insulin to a weight-based regimen for control of hyperglycemia, using the primary outcome length of stay, in non-ICU inpatients within the northeast region of the VA. All non-ICU patients who require in-hospital insulin therapy are eligible for the trial, and the VA's automated systems will be used to assess eligibility and present the possibility of randomization to the clinician at the point of care. Clinicians will indicate their approval for informed consent to be obtained by study staff. Adaptive randomization will assign up to 3000 patients, preferentially to the currently 'winning' strategy, and all care will proceed according to usual practices. Based on a Bayesian stopping rule, the study has acceptable frequentist operating characteristics (Type I error 6%, power 86%) against a 12% reduction of median length of stay from 5 to 4.4 days. The adaptive stopping rule promotes implementation of a successful treatment strategy. LIMITATIONS: Despite clinical equipoise, individual healthcare providers may have strong treatment preferences that jeopardize the success and implementation of the trial design, leading to low rates of randomization. Unblinded treatment assignment may bias results. In addition, generalization of clinical results to other healthcare systems may be limited by differences in patient population. Generalizability of the POC-CT method depends on the level of informatics and statistics infrastructure available to a healthcare system. CONCLUSIONS: The methods proposed will demonstrate outcome-based evaluation of control of hyperglycemia in hospitalized veterans. By institutionalizing a process of statistically sound and efficient learning, and by integrating that learning with automatic implementation of best practice, the participating VA Healthcare Systems will accelerate improvements in the effectiveness of care.

Comparative effectiveness research and medical informatics.

As is the case for environmental, ecological, astronomical, and other sciences, medical practice and research finds itself in a tsunami of data. This data deluge, due primarily to the introduction of digitalization in routine medical care and medical research, affords the opportunity for improved patient care and scientific discovery. Medical informatics is the subdiscipline of medicine created to make greater use of information in order to improve healthcare. The 4 areas of medical informatics research (information access, structure, analysis, and interaction) are used as a framework to discuss the overlap in information needs of comparative effectiveness research and potential contributions of medical informatics. Examples of progress from the medical informatics literature and the Veterans Affairs Healthcare System are provided.

Adverse events associated with testosterone administration.

BACKGROUND: Testosterone supplementation has been shown to increase muscle mass and strength in healthy older men. The safety and efficacy of testosterone treatment in older men who have limitations in mobility have not been studied. METHODS: Community-dwelling men, 65 years of age or older, with limitations in mobility and a total serum testosterone level of 100 to 350 ng per deciliter (3.5 to 12.1 nmol per liter) or a free serum testosterone level of less than 50 pg per milliliter (173 pmol per liter) were randomly assigned to receive placebo gel or testosterone gel, to be applied daily for 6 months. Adverse events were categorized with the use of the Medical Dictionary for Regulatory Activities classification. The data and safety monitoring board recommended that the trial be discontinued early because there was a significantly higher rate of adverse cardiovascular events in the testosterone group than in the placebo group. RESULTS: A total of 209 men (mean age, 74 years) were enrolled at the time the trial was terminated. At baseline, there was a high prevalence of hypertension, diabetes, hyperlipidemia, and obesity among the participants. During the course of the study, the testosterone group had higher rates of cardiac, respiratory, and dermatologic events than did the placebo group. A total of 23 subjects in the testosterone group, as compared with 5 in the placebo group, had cardiovascular-related adverse events. The relative risk of a cardiovascular-related adverse event remained constant throughout the 6-month treatment period. As compared with the placebo group, the testosterone group had significantly greater improvements in leg-press and chest-press strength and in stair climbing while carrying a load. CONCLUSIONS: In this population of older men with limitations in mobility and a high prevalence of chronic disease, the application of a testosterone gel was associated with an increased risk of cardiovascular adverse events. The small size of the trial and the unique population prevent broader inferences from being made about the safety of testosterone therapy. (ClinicalTrials.gov number, NCT00240981.)

Evaluation of a generalizable approach to clinical information retrieval using the automated retrieval console (ARC).

Reducing custom software development effort is an important goal in information retrieval (IR). This study evaluated a generalizable approach involving with no custom software or rules development. The study used documents "consistent with cancer" to evaluate system performance in the domains of colorectal (CRC), prostate (PC), and lung (LC) cancer. Using an end-user-supplied reference set, the automated retrieval console (ARC) iteratively calculated performance of combinations of natural language processing-derived features and supervised classification algorithms. Training and testing involved 10-fold cross-validation for three sets of 500 documents each. Performance metrics included recall, precision, and F-measure. Annotation time for five physicians was also measured. Top performing algorithms had recall, precision, and F-measure values as follows: for CRC, 0.90, 0.92, and 0.89, respectively; for PC, 0.97, 0.95, and 0.94; and for LC, 0.76, 0.80, and 0.75. In all but one case, conditional random fields outperformed maximum entropy-based classifiers. Algorithms had good performance without custom code or rules development, but performance varied by specific application.

Association of elevated alanine aminotransferase with BMI and diabetes in older veteran outpatients.

AIMS: We determined age-stratified associations of elevated alanine aminotransferase (ALAT), as a likely indicator of non-alcoholic steatonecrosis (NASH), with body mass index (BMI) and diabetes among older male, veteran outpatients. METHODS: Administrative files at three, urban Veteran Administration Healthcare Systems were surveyed. 47.916 male outpatients, mean age 69.2 years, without ICD-9CM documented liver diseases or alcohol- or drug-abuse, had the highest of one or more ALAT results analyzed. Logistic regression determined age-stratified predictors of elevated ALAT among groups <50 ("younger-aged"), 50-70 ("middle-aged") and >70 ("older-aged"). BMI, diabetes and "statin" use were predictors. RESULTS: 1272 (2.7%) patients had their highest ALAT greater than the upper limit of normal (mean=98.4U/L). 4.9% of all diabetics and 15.2, 6.1 and 3.4% of the "younger-aged", middle-aged" and "older-aged" diabetics, respectively, had elevated ALATs. BMI values from 25 to <30 and 30 to 45, compared with those from >20 to <25, predicted elevated ALAT in the two younger groups. In contrast, among "older-aged" patients, diabetes independently predicted elevated ALAT, whereas BMI did not. CONCLUSIONS: The cross-sectional findings support previous epidemiologic data associating ALAT elevations with obesity and diabetes, and provide new evidence that in elderly male patients, ALAT elevations, absent other known liver diseases, may selectively associate with diabetes.

The association between statins and cancer incidence in a veterans population.

BACKGROUND: Meta-analyses of trials of 3-hydroxy-3-methyl-glutaryl-CoA reductase inhibitors or statins for cardiovascular disease prevention have failed to show any statistically significant benefit of statins for cancer prevention. However, these trials included relatively young participants, who develop few cancers, and their follow-up periods may have been too short to detect an association between statin use and cancer incidence. We investigated this association in a population of veterans. METHODS: We identified patients using antihypertensive medications but no cholesterol-lowering medications (n = 25,594) and patients using statins (n = 37,248) who were enrolled in the Veterans Affairs New England Healthcare System between January 1, 1997, and December 31, 2005. Age- and multivariable-adjusted Cox proportional hazards models were used to calculate the hazard ratio (HR) and its 95% confidence interval (CI) for cancer incidence, excluding nonmelanoma skin cancer, among patients taking statins compared with patients taking antihypertensive medications and among patients grouped by statin dose (as equivalent simvastatin dose). All statistical tests were two-sided. RESULTS: The absolute incidence of total cancers was 9.4% among statin users and 13.2% among nonusers (difference = 3.8%, 95% CI = 3.3% to 4.3%, P(difference) < .001). Statin users had a statistically significant lower risk for total cancer than nonusers after adjustment for age (HR = 0.76, 95% CI = 0.73 to 0.80) and multiple potential confounders (HR = 0.74, 95% CI = 0.70 to 0.78). After multivariable adjustment, a statistically significantly decreased risk of all cancers was also associated with increasing statin use (P(trend) < .001). CONCLUSIONS: Patients using statins may be at lower risk for developing cancer. Additional observational studies and randomized trials of statins for cancer prevention are warranted.

Quetiapine versus trazodone in reducing rehospitalization for alcohol dependence: a large data-base study.

Clinicians commonly use nonindicated (or "off-label") medications for the treatment of alcoholism, although there is no clear evidence to support this practice. Quetiapine and trazodone are 2 medications frequently used in this manner, especially to treat sleep disturbance associated with alcohol withdrawal. Using national administrative data from the Department of Veterans Affairs, we compared the differences among these medications in time to rehospitalization after discharge from an index hospitalization for alcohol dependence. Our outcome measure was the difference in time (weeks) to rehospitalization for patients given medications in 1 of 4 groups: quetiapine alone (median, 6.1 weeks); trazodone alone (median, 10.1 weeks); quetiapine combined (median, 7.1 weeks); and trazodone combined (median, 10.3 weeks) with other frequently used psychotropic drugs. Differences between groups were examined with Cox's proportional hazards regression using trazodone in combination with other medications as the reference category. Crude hazard ratios were determined first and then adjusted for covariates. There was no difference in the risk of rehospitalization when patients using quetiapine in combination were compared with the reference category (hazard ratio [HR] = 1.08; confidence interval [CI], 0.99-1.17; P < 0.0936). In contrast, when quetiapine was used alone there was a significantly higher risk of rehospitalization (HR = 1.22; CI = 1.06-1.41; P < 0.005). When trazodone was used alone there was no difference in risk in the unadjusted analysis (HR = 1.05; CI = 0.96-1.14; P < 0.3076); however, the HR increased to a significant level after adjusting for covariates (HR, 1.14; CI = 1.05-1.24; P < 0.0029) and for those with 2+ previous discharges (HR = 1.25; CI = 1.14-1.37; P < 0.0001). These findings suggest the need for additional studies to better understand what symptoms of alcoholism benefit from the use of these medications, at what dose levels, and with what other medication combinations.

Long-term, high-dose benzodiazepine prescriptions in veteran patients with PTSD: influence of preexisting alcoholism and drug-abuse diagnoses.

Databases from the New England Veterans Integrated Service Network were analyzed to determine factors associated with long-term, high-dose anxiolytic benzodiazepine prescriptions dispensed to patients with posttraumatic stress disorder (PTSD) and existing alcoholism and/or drug abuse diagnoses. Among 2,183 PTSD patients, 234 received the highest 10% average daily doses for alprazolam, clonazepam, diazepam, or lorazepam, doses above those typically recommended. Highest doses were more commonly prescribed to patients with existing drug abuse diagnoses. Among patients with PTSD and alcoholism, younger age, drug abuse, and concurrent prescriptions for another benzodiazepine and oxycodone/acetaminophen independently predicted high doses. Results indicate that for veteran patients with PTSD, alcoholism alone is not associated with high-dose benzodiazepines, but existing drug abuse diagnoses do increase that risk.

Health care utilization and receipt of cholesterol testing by veterans with and those without mental illness.

OBJECTIVE: We examined the relationship between mental illness, health care utilization and rates of cholesterol testing. METHODS: We conducted a retrospective cohort study using Veterans Affairs (VA) administrative data on 64,490 United States veterans who used VA New England Health Care System outpatient services between January 1998 and June 2001. A total of 10,100 veterans (15.7%) had a mental illness treated with medication. We examined the interaction between mental illness and outpatient service utilization with respect to the likelihood of receiving a cholesterol test, adjusting for major demographic and clinical covariates. RESULTS: Among veterans using VA outpatient services infrequently, those with mental illness were less likely than non-mentally ill control subjects to receive a cholesterol test during the study period (first quartile adjusted OR=0.45, 95% CI=0.37-0.54; second quartile adjusted OR=0.50, 95% CI=0.45-0.57). Mentally ill subjects with more frequent utilization of VA services were as likely as (third quartile adjusted OR=1.01, 95% CI=0.91-1.13) or more likely than (fourth quartile adjusted OR=2.73, 95% CI=2.46-3.03) non-mentally ill subjects to receive cholesterol testing. CONCLUSIONS: Mental illness was associated with a lower likelihood of cholesterol testing in subjects who used fewer VA outpatient services. The observed disparity attenuated at higher levels of service utilization.