Large Electronic Medical Record Research Articles

Comparing the Performance of Charlson and Elixhauser Comorbidity Indices to Predict In-Hospital Mortality Among a Chinese Population.

ObjectiveEarlier comorbidity measures have been developed or validated using the North American population. This study aims to compare five Charlson or Elixhauser comorbidity indices to predict in-hospital mortality using a large electronic medical record database from Shanxi, China.MethodsUsing the primary diagnosis code and surgery procedure codes, we identified four hospitalized patient cohorts, hospitalized between 2013 and 2017, in Shanxi, China, as follows: congestive heart failure (CHF, n=41,577), chronic renal failure (CRF, n=40,419), diabetes (n=171,355), and percutaneous coronary intervention (PCI, n=39,097). We used logistic regression models and c-statistics to evaluate the in-hospital mortality predictive performance of two multiple comorbidity indicator variables developed by Charlson in 1987 and Elixhauser in 1998 and three single numeric scores by Quan in 2011, van Walraven in 2009, and Moore 2017.ResultsElixhauser comorbidity indicator variables had consistently higher c-statistics (0.824, 0.843, 0.904, 0.853) than all other four comorbidity measures, across all four disease cohorts. Moore’s comorbidity score outperformed the other two score systems in CHF, CRF, and diabetes cohorts (c-statistics: 0.776, 0.832, 0.869), while van Walraven’s score outperformed all others among PCI patients (c-statistics: 0.827).ConclusionElixhauser comorbidity indicator variables are recommended, when applied to large Chinese electronic medical record databases, while Moore’s score system is appropriate for relatively small databases.

From benefits idealisation to value optimisation: application in the digital health context.

Objective This study investigated evidence for the approach known as 'benefits management' (BM) used in many digital hospital initiatives. Methods A qualitative narrative overview was conducted on the BM literature and compared with a qualitative systematic overview of electronic medical record (EMR) implementation literature. Results Twenty-five articles on BM and 12 literature reviews on EMR implementation were examined. The BM approach does not have strong support in the literature and does not support all the needs of large EMR implementations. Conclusion The current BM approach provides an inadequate basis for managing and reporting on the outcomes that ensue from a digital hospital initiative. A shift is needed from benefits idealisation to value optimisation. What is known about the topic? Health services are under increasing pressure to demonstrate that the benefits anticipated from digital health investments have been realised. What does this paper add? This paper informs the practice of benefits governance in EMR implementations. The results reveal inadequacies in current BM models and practice that are currently enshrined in policy despite a lack of evidence. What are the implications for practitioners? Health service leaders must be willing to question the governance of benefits from health service transformations using more evidence-based approaches to increase the value obtained from investments in digital transformation.

The characteristics of patients with possible familial hypercholesterolemia-screening a large payer/provider healthcare delivery system.

Familial hypercholesterolemia (FH) is an under-diagnosed condition. We applied standard laboratory criteria across a large longitudinal electronic medical record database to describe cross-sectional population with possible FH. A cross-sectional study of Clalit Health Services members. Subjects who met the General Population MED-PED laboratory criteria, excluding: age <10 years, documentation of thyroid, liver, biliary or autoimmune diseases, a history of chronic kidney disease stage 3 or greater, the presence of urine protein >300mg/l, HDL-C>80mg/dl, active malignancy or pregnancy at the time of testing were considered possible FH. Demographic and clinical characteristics are described at time of diagnosis and at a single index date following diagnosis to estimate the burden on the healthcare system. The patient population is also compared to the general population. The study cohort included 12494 subjects with out of over 4.5 million members of Clalit Health Services. The estimated prevalence of FH in Israel was found to be 1:285. These patients are notably positive for, and have a family history of, cardiovascular disease and risk factors. For most of them the LDL-C levels are not controlled, and only a quarter of them are medically treated. By using the modified MED-PED criteria in a large electronic database, patients with possible FH can be identified enabling early intervention and treatment.

A Canadian strategy for surgical quality improvement

The Canadian Association of General Surgeons (CAGS) Board of Directors hosted a symposium to develop a Canadian strategy for surgical quality and safety at its mid-term meeting on Feb. 24, 2018. The following 6 principles outline the consensus of this symposium, which included diverse stakeholders and surgeon leaders across Canada: 1) a Canadian quality-improvement strategy for surgery is needed; 2) quality improvement requires continuous, active and intentional effort; 3) outcome measurement alone will not drive improvement; 4) increased focus on standardization and process improvement is necessary; 5) new, large electronic medical record systems pose challenges as well as benefits in Canadian hospitals; and 6) surgeons in remote and rural hospitals must be engaged using tailored approaches.

Comparative United States autoimmune disease rates for 2010-2016 by sex, geographic region, and race.

AIDs may disproportionately impact specific racial groups, but autoimmune (AID) prevalence information by minority racial group is sparse for many AIDs. The objective of this analysis was to supplement previously published AID prevalence rates by providing information on race rate ratios (minority race populations compared to Caucasian populations) in the United States. Preliminary to estimating race rate ratios, contemporary US-specific, health care utilization-based AID prevalence rates and female-to-male ratios were estimated and compared to previously published AID prevalence rates. We used a large national electronic medical record database of 52 million individuals to estimate age-adjusted direct standardized rates for 22 AIDs for 2010 through 2016 by gender, race, and US census division. These were compared to previously published estimates. Female-to-male ratios were comparable with published studies. Almost all observed Multiracial AID rates were significantly higher than Caucasian rates, as well as 9 of 22 AID rates observed among Native Americans and 8 of 22 AID rates estimated among African-American patients. Regional variation was noted: highest African-American systemic lupus erythematosus rates were observed in the West North Central and South Atlantic divisions, highest African-American multiple sclerosis rates in the South Atlantic and Pacific divisions, and highest Native American rheumatoid arthritis rates in the West North Central, Mountain, and Pacific divisions. Substantial AID heterogeneity exists by race and by geographic area. An important research area is further exploring factors related to heterogeneity such as potential interactions between genetic susceptibility and environmental factors.

A Discrete Event Simulation Model to Evaluate the Impact of Treatment Sequences on Long-Term Patient Outcomes in Multiple Myeloma (MM)

Purpose: Many new therapies have been approved for the treatment of multiple myeloma (MM) in recent years. In June 2019, daratumumab (Darzalex®), a CD38-directed monoclonal antibody, in combination with lenalidomide and dexamethasone (DRd), received FDA approval for the treatment of newly diagnosed MM (NDMM) patients ineligible for stem cell transplantation. This approval provides NDMM patients with a new option as first-line therapy, however, with many options for MM treatment available, identification of the optimal sequence of treatments becomes of critical importance to healthcare providers and payers. Objective: To develop a treatment sequencing discrete event simulation (DES) model that compares unique treatment sequences and estimates the impact of first-line DRd on longer-term outcomes (i.e. survival through multiple lines of therapy) in a general population of MM patients. Methods: A DES model of MM treatment sequences was developed. FDA approved and NCCN recommended non-transplant therapies were included. Treatment efficacy was estimated by digitizing and extrapolating progression-free and overall survival (OS) Kaplan-Meier (KM) data, using established methods (Guyot et al., 2012). Five common parametric functions were fit, and the curve-of-best-fit was determined by standard statistics (i.e., Akaike &amp; Bayesian information criterion) and visual inspection. The model was developed to allow patients to go through up to 6 lines of treatment. The time of progression and death events were randomly assigned based on the estimated parametric functions. All-cause mortality data (US Census 2016) were considered in the simulation of survival. Where data were not available, previous therapies were assumed not to impact the efficacy of subsequent therapies. Model survival predictions were validated against published real-world estimates of survival from: 1) a retrospective analysis of a large American electronic medical record database, Humedica (Hari et al., 2018) and 2) a retrospective analysis of the Dutch PHAROS registry (Verelst et al., 2017). The Humedica and PHAROS datasets reported on the effects of three and four lines of treatment, respectively, starting with newly diagnosed patients. From the Humedica dataset, first-line therapy consisted of proteasome inhibitor- (PI), immunomodulatory- (IMiD), and PI + IMiD-based regimens in 42%, 34% and 18% of patients, respectively, with other regimens used in 7% of cases. Treatment was primarily doublets (63%) rather than triplets (37%). From the PHAROS registry, first-line therapy was primarily thalidomide-based (66%), bortezomib-based (15%), or lenalidomide-based (7%). Statistically, validation was quantified through the calculation of the mean absolute percent error (MAPE), a commonly used statistic to assess the accuracy of forecasts. To estimate the impact of first-line DRd, a hypothetical analysis was simulated wherein all first-line treatments described within both the Humedica and PHAROS datasets were replaced with DRd. Following DRd, patients received subsequent treatment as described within those datasets. Results: Restriction of the modeled treatment algorithm to treatment patterns described within the Humedica database and PHAROS registry led to valid simulated OS estimates that were in line with the published data (MAPE values of 9% and 19% for Humedica and PHAROS, respectively). Utilization of DRd as first-line treatment was estimated to increase OS in scenario analyses (Table 1). The DES model estimated 5-year survival rates for frontline DRd with subsequent therapy based upon the Humedica and PHAROS datasets were 62% and 57%, respectively. In contrast, the estimated 5-year survival for the Humedica and PHAROS datasets without DRd frontline was 40% and 26% respectively, highlighting a 22-31% incremental survival benefit due to DRd at 5-years. Conclusions: The MM DES model validated well to the published estimates of real-world OS. Preliminary results indicate that first-line treatment with DRd may improve OS in MM compared with historical treatment sequences, that include PI+IMiD based triplets as first-line treatment. The model is being further developed to describe the impact of sequencing other MM treatment regimens to estimate the optimal treatment pathways. Disclosures Liu: Janssen Scientific Affairs, LLC: Employment, Equity Ownership. Moldaver:Janssen Scientific Affairs, LLC: Consultancy. Zhu:Janssen Scientific Affairs, LLC: Other: contractor for Janssen. Zhou:Janssen Scientific Affairs, LLC: Consultancy. Maiese:Janssen: Employment, Equity Ownership. Hollmann:Janssen Scientific Affairs, LLC: Consultancy.

PDB2 USING REAL WORLD DATA TO EXAMINE TRIGLYCERIDES, LDL-CHOLESTEROL, AND HDL-CHOLESTEROL AS PREDICTORS OF PROGRESSION FROM NONALCOHOLIC STEATOHEPATITIS TO HEPATOCELLULAR CANCER AND LIVER CIRRHOSIS

The aims of this study were to (1) examine changes in triglycerides, low-density lipoprotein (LDL) cholesterol, and high-density lipoprotein (HDL) cholesterol after a diagnosis of nonalcoholic steatohepatitis (NASH), and (2) measure the incidence of liver cirrhosis (LC) and hepatocellular cancer (HCC) by lipid type. Patients with a laboratory measure for triglycerides, LDL-, or HDL-cholesterol in the year prior to their first diagnosis of NASH (ICD-10 code K75.81; defined as baseline period) were identified in a large electronic medical record network. Mean values for each lipid were assessed in the 1st, 2nd, and 3rd year after the first NASH diagnosis. LC and HCC incidence were measured in patients with consistently abnormal and normal baseline values. Risk ratios (95% CIs) compared triglyceride (150+ vs <150 mg/dL), HDL (<60 vs 60+ mg/dL), and LDL (130+ vs <130 mg/dL)] values, adjusting for confounding using a 1:1 matched greedy-nearest-neighbor propensity score model. Patients were 57±13 years old and 45% male at baseline (triglycerides: N=12,070; LDL: N=12,042; and HDL: N=13,124). After the incident diagnosis of NASH, the mean triglyceride values increased monotonically (baseline:176 mg/dL; 3rd year: 185 mg/dL), LDL values decreased monotonically (baseline: 96.8 mg/dL; 3rd year: 90 mg/dL), and HDL values remained stable. After adjusting for co-morbidities and baseline medications, patients with abnormal triglycerides were 1.6 (1.1-2.4) times as likely to develop LC. Patients with abnormal HDL, were 1.6 (0.7-3.3) times as likely to develop HCC. Other models did not predict progression. Mean triglyceride values increased in the 3-years following the incident NASH diagnosis and elevated triglycerides at baseline predicted progression to LC. Although not statistically significant, patients with abnormal HDL had an increased risk of HCC. Robust laboratory data provides real-world-evidence to support studies showing triglyceride values to be an important bio-marker for disease progression and interventions in NASH patients.

Measuring the Effectiveness of Safety Warnings on the Risk of Stroke in Older Antipsychotic Users: A Nationwide Cohort Study in Two Large Electronic Medical Records Databases in the United Kingdom and Italy.

Safety warnings relating to antipsychotic-associated stroke among older persons in the UK and Italy were issued.However, the impact of these safety warnings on stroke risk has not been measured to date. The aim of this study was to measure the change in stroke incidence after two safety warnings in both the UK and Italy. A cohort study was conducted using electronic medical records representative of the UK (The Health Improvement Network) and Italy (Health Search-IQVIA Health LPD), containing data on 11million and 1 million patients, respectively. After each drug safety warning, elderly antipsychotic new initiators were propensity-score matched 1:1:1 on antipsychotic initiators before any safety warning. Stroke incidence within 6months of antipsychotic initiation, using an intention-to-treat approach, was the main outcome. In the UK and Italy, 6342 and 7587 elderly antipsychotic initiators were identified, respectively. A 42% stroke incidence reduction was seen in the UK after the first safety warning [42.3 (95% confidence interval (CI) 35.2-50.8) vs. 24.4 [95% CI 19.0-31.2] events per 1000 person-years (PYs)], while there was a 60% stroke incidence reduction after the second warning (16.9 [95% CI 12.2-23.4] events per 1000 PYs)compared to before the first warning. There was no significant reduction in stroke incidence in Italy. Antipsychotic safety warnings were followed by a reduction in stroke incidence among older antipsychotic users in the UK, but not Italy.

Machine learning methods for developing precision treatment rules with observational data

Clinical trials have identified a variety of predictor variables for use in precision treatment protocols, ranging from clinical biomarkers and symptom profiles to self-report measures of various sorts. Although such variables are informative collectively, none has proven sufficiently powerful to guide optimal treatment selection individually. This has prompted growing interest in the development of composite precision treatment rules (PTRs) that are constructed by combining information across a range of predictors. But this work has been hampered by the generally small samples in randomized clinical trials and the use of suboptimal analysis methods to analyze the resulting data. In this paper, we propose to address the sample size problem by: working with large observational electronic medical record databases rather than controlled clinical trials to develop preliminary PTRs; validating these preliminary PTRs in subsequent pragmatic trials; and using ensemble machine learning methods rather than individual algorithms to carry out statistical analyses to develop the PTRs. The major challenges in this proposed approach are that treatment are not randomly assigned in observational databases and that these databases often lack measures of key prescriptive predictors and mental disorder treatment outcomes. We proposed a tiered case-cohort design approach that uses innovative methods for measuring and balancing baseline covariates and estimating PTRs to address these challenges.

An Application of Machine Learning in IVF: Comparing the Accuracy of Classification Alogithims for the Prediction of Twins

Background: Clinical decision-making dilemmas are particularly notable in IVF practice, given that large datasets are often generated which enable clinicians to make predictions that inform treatment choices. This study applied machine learning by using IVF data to determine the risk of twins when two or more embryos are available for transfer. While most classifiers are able to provide estimates of accuracy, this study went further by comparing classifiers both by accuracy and Area Under the Curve (AUC).Methods: Study data were derived from a large electronic medical record system that is utilized by over 140 IVF clinics and contained 135,000 IVF cycles. The dataset was reduced from 88 variables to 40 and included only those cycles of IVF where two or more blastocyst embryos were created. The following classifiers were compared in terms of accuracy and AUC: a generalized linear model, linear discriminant analysis, quadratic discriminant analysis, K-nearest neighbors, support vector machine, random forests, and boosting. A stacking ensemble learning algorithm was also applied in order to use predictions from classifiers to create a new model.Results: While the ensemble classifier was the most accurate, none of the classifiers predominated as being significantly superior to other classifiers. Findings indicated that boosting methods for classifiers performed poorly; logistic and linear discriminant analysis classifiers performed better than the quadratic discriminant analysis classifier, and the support vector machine performed almost as well as the tree classifier. AUC results were consistent with the comparisons for accuracy. External validation was also performed using a different dataset containing 588 observations. All models performed better using the external validation dataset, with the random forest classifier performing markedly better than any other classifier.Conclusions: These results support the impression that big data can be of value in the clinical decision-making process; but that no single statistical algorithm provides maximum accuracy for all databases. Therefore, different datasets will require investigation in order to determine which algorithms are the most accurate for a particular set of data. These findings underscore the premise that clinicians with access to large amounts of data can use advanced predictive analytic models to create robust clinical information of vital importance for patient care.

Polygenic risk score identifies associations between sleep duration and diseases determined from an electronic medical record biobank.

We aimed to detect cross-sectional phenotype and polygenic risk score (PRS) associations between sleep duration and prevalent diseases using the Partners Biobank, a hospital-based cohort study linking electronic medical records (EMR) with genetic information. Disease prevalence was determined from EMR, and sleep duration was self-reported. A PRS for sleep duration was derived using 78 previously associated SNPs from genome-wide association studies (GWAS) for self-reported sleep duration. We tested for associations between (1) self-reported sleep duration and 22 prevalent diseases (n = 30 251), (2) the PRS and self-reported sleep duration (n = 6903), and (3) the PRS and the 22 prevalent diseases (n = 16 033). For observed PRS-disease associations, we tested causality using two-sample Mendelian randomization (MR). In the age-, sex-, and race-adjusted model, U-shaped associations were observed for sleep duration and asthma, depression, hypertension, insomnia, obesity, obstructive sleep apnea, and type 2 diabetes, where both short and long sleepers had higher odds for these diseases than normal sleepers (p < 2.27 × 10-3). Next, we confirmed associations between the PRS and longer sleep duration (0.65 ± 0.19 SD minutes per effect allele; p = 7.32 × 10-04). The PRS collectively explained 1.4% of the phenotypic variance in sleep duration. After adjusting for age, sex, genotyping array, and principal components of ancestry, we observed that the PRS was also associated with congestive heart failure (CHF; p = 0.015), obesity (p = 0.019), hypertension (p = 0.039), restless legs syndrome (RLS; p = 0.041), and insomnia (p = 0.049). Associations were maintained following additional adjustment for obesity status, except for hypertension and insomnia. For all diseases, except RLS, carrying a higher genetic burden of the 78 sleep duration-increasing alleles (i.e. higher sleep duration PRS) associated with lower odds for prevalent disease. In MR, we estimated causal associations between genetically defined longer sleep duration with decreased risk of CHF (inverse variance weighted [IVW] OR per minute of sleep [95% CI] = 0.978 [0.961-0.996]; p = 0.019) and hypertension (IVW OR [95% CI] = 0.993 [0.986-1.000]; p = 0.049), and increased risk of RLS (IVW OR [95% CI] = 1.018 [1.000-1.036]; p = 0.045). By validating the PRS for sleep duration and identifying cross-phenotype associations, we lay the groundwork for future investigations on the intersection between sleep, genetics, clinical measures, and diseases using large EMR datasets.

Baseline Characteristics of Patients with Paroxysmal Nocturnal Hemoglobinuria Identified in the Department of Defense Database

Background: Rare diseases are often characterized by misdiagnosis resulting in delays in critical and potentially lifesaving treatment. Awareness of first signs and symptoms of rare diseases can provide clinical evidence for early and accurate diagnosis. Medical records are the primary source of clinical information from first signs and symptoms to key clinical disease related events, yet access to records can be difficult and time consuming. We conducted a study using a large electronic medical record (EMR) and claims database to identify and describe characteristics of Paroxysmal Nocturnal Hemoglobinuria (PNH) patients at the time of their PNH diagnosis. We present baseline findings of the PNH population identified in the Department of Defense (DOD) healthcare system, where all clinical details were readily available. The DOD healthcare system is a US-based, longitudinal EMR and claims database with health information on approximately 10 million active beneficiaries throughout the country.Methods: We identified all people in the DOD database from January 01, 2007 through May 31, 2017 who had an ICD-10-CM code for PNH or a NDC or HCPCS code for eculizumab, the drug used to treat PNH. There is no ICD-9-CM code for PNH. Cases had no other indication for eculizumab use and had to have appropriate symptoms, comorbidities or lab results to be considered a case. We reviewed the electronic record for each patient and classified each as definite/likely, probable/possible or unlikely PNH based on attributes of the available health data including codes for eculizumab, flow cytometry, lab results, hemoglobinuria, aplastic anemia, pancytopenia and other comorbidities. Our ruling was validated by medical record review by a clinical expert for all equivocal cases and a sample of definite/likely cases. Individuals classified as unlikely PNH were excluded. Patient characteristics at PNH diagnosis are presented using descriptive statistics.Results: We identified 73 PNH patients (55% female) after review of all available electronic data; 41 had a diagnosis of PNH and received eculizumab, 17 had a diagnosis of PNH only, and 15 received eculizumab only. From these, 61 patients were determined to have definite/likely PNH and 12 had probable/possible PNH. The use of eculizumab ranged from 0 to 378 (median=18) prescriptions, dependent in part on the amount of follow-up in the database. There were 19 patients (26%) who had prevalent PNH and 54 (74%) who were newly diagnosed with PNH during the study period. Among newly diagnosed patients, the median age at PNH diagnosis was 46 years (range 13 - 86). Most patients had codes for hemoglobinuria at or before the PNH diagnosis date (69%), and many had codes for aplastic anemia (46%), thrombocytopenia (56%), and pancytopenia (28%). At any time after the PNH diagnosis, 85% had unspecified anemia, 52% had aplastic anemia, 53% had thrombocytopenia, and 33% had pancytopenia codes in their records. See table.Conclusion: From a population of around 10 million actively enrolled patients, across a 10-year study period, we identified 73 patients with PNH. Data from these patients will be used to identify patterns of health encounters leading up to diagnosis and relevant outcomes following diagnosis. This information can be used to help diagnose other patients with this rare disease and to improve their medical outcomes.Disclaimer Statement: Research data were derived from an approved Naval Medical Center, Portsmouth, VA IRB protocol (NMCP.2017.0080). The views expressed in this abstract are those of the authors and do not necessarily reflect the official policy or position of the Department of the Navy, Department of Defense or the United States Government. Copyright Notice: CAPT Brian Feldman is a military service member. This work was prepared as part of his official duties. Title 17 U.S.C. 105 provides that ‘Copyright protection under this title is not available for any work of the United States Government.‘Title 17 U.S.C. 101 defines a United States Government work as a work prepared by a military service member or employee of the United States Government as part of that person's official duties. DisclosuresUlcickas Yood:Alexion Pharmaceuticals, Inc.: Other: Employee of EpiSource, LLC, which was contracted by Alexion Pharmaceuticals, Inc. EpiSource had the final decision on content. . Jick:Alexion Pharmaceuticals, Inc.: Other: Employee of the Boston Collaborative Drug Surveillance Program, which was contracted and paid by Alexion Pharmaceuticals, Inc. to work on a study of PNH using DOD data. . Vasilakis-Scaramozza:Alexion Pharmaceuticals, Inc.: Other: Employee of the Boston Collaborative Drug Surveillance Program, which was contracted and paid by Alexion Pharmaceuticals, Inc. to work on a study of PNH using DOD data.. Donato:Alexion Pharmaceuticals Inc: Employment, Equity Ownership. Tomazos:Alexion Pharmaceuticals, Inc.: Employment, Equity Ownership. L'Italien:Alexion Pharmaceuticals, Inc.: Equity Ownership, Other: Former employee and current stockholder of Alexion Pharmaceuticals, Inc. . Sicignano:Alexion Pharmaceuticals, Inc.: Other: Employee of Health ResearchTx, which has a business relationship with Alexion Pharmaceuticals, Inc.. Feldman:Alexion Pharmaceuticals, Inc.: Other: Employee, Department of Navy, United States Government..

Current Real-World Treatment Patterns and Outcomes in Patients with Relapsed/Refractory Multiple Myeloma

Introduction: Limited real-world data exist on treatments for relapsed/refractory multiple myeloma (RRMM) since the approval of several new agents. This study assessed treatment patterns and outcomes of patients with RRMM receiving ≥2 lines of therapy in US community oncology practices.Methods: A chart review was conducted in patients ≥18 years with MM diagnosed January 1, 2011-May 31, 2017, from a large electronic medical record database. Patient data were examined from the date of initiation of first-line therapy (1LT) for MM until death, loss to follow-up, or study end date, whichever was earliest. This study was hypothesis generating, thus no statistical tests were performed. Descriptive statistics were used to describe baseline demographic/clinical characteristics, treatment patterns, median progression-free survival (mPFS) and median overall survival (mOS) for the overall RRMM population. Patients were then stratified into “older” vs. “newer” treatment cohorts based on whether the drugs used in each line of therapy were approved before, or after 2013.Results: Of 1005 charts reviewed, 456 patients had received ≥2 lines of therapy and were included in the chart review study (median age at diagnosis: 70.4 years; females: 39.5%, males 60.5%; bone involvement at diagnosis: 66.0%; International Staging System stage within 1 month of diagnosis: I 28.7%, II 27.9%, III 43.4%; 183 (40.1%) patients received 3LT, 75 (16.4%) 4LT, and 29 (6.4%) 5LT. 1LT was dominated by bortezomib (BTZ), lenalidomide (LEN), and the combination of the two, with 93.3% of patients using these agents as 1LT and 69.8% of patients as 2LT. In 3LT and beyond, there was greater use of newly approved drugs (approved since 2013) compared with 1LT and 2LT; pomalidomide (14.8% 3LT, 17.3% 4LT, 17.2% 5LT), carfilzomib (19.1% 3LT, 13.3% 4LT, 17.2% 5LT), elotuzumab (2.2% 3LT, 4.0% 4LT, 3.4% 5LT), panobinostat (0% 3LT, 2.7% 4LT, 3.4% 5LT), daratumumab (6.6% 3LT, 13.3% 4LT, 17.2% 5LT), and ixazomib (0% 3LT, 1.3% 4LT, 3.4% 5LT) with the exception of the BTZ/LEN combination. However, patients receiving either BTZ, LEN, or both in combination as 1LT or 2LT often received the agent(s) as re-treatment in lines 2-6 (46.2%-55.6%). Median time on treatment decreased from 7.5 months in 1LT to ≤2.3 months in 4LT and 5LT, and median treatment-free intervals decreased from 1.6 months between 1LT and 2LT to 0.5 months between 4LT and 5LT. The most common reason for discontinuation was disease progression and drug toxicity/intolerability.The most commonly reported adverse events (AEs) for all lines of therapy were fatigue (71.6%-78.3%), bone pain (38.5%-69.1%) and anemia (53.8%-69.3%). AEs were generally constant across lines of therapy, except for bone pain, anemia, and neuropathy, which broadly decreased with increasing line of therapy. Overall, mPFS ranged from 12.0 months in 1LT to 3.5 months in 5LT and mOS ranged from 48.2 months in 1LT to 5.8 months in 5LT (Figure 1). A trend for increased mPFS and mOS with newer vs. older drugs was observed across treatment lines. The magnitude of the “new” treatment benefit on mPFS was pronounced in 1LT.Conclusions: Forty percent of patients received therapy beyond 2lines demonstrating a great unmet need in the treatment of RRMM. While BTZ and LEN were predominant in 1st and 2nd lines, substantial fragmentation was seen in ≥3 line, highlighting the lack of defined treatment pathways for these patients. The results further indicated that treatments beyond 2LT offer shorter benefit as disease progresses; median time on treatment and mPFS decreased as treatment line increased. Also, median PFS and OS in this real-world analysis were also slightly lower than that observed in recent clinical trials of novel agents, which is consistent with other real-world studies. While there remains a need to replicate these results within a larger dataset where statistical comparisons could be made and confounding factors controlled for, the trends observed in this study suggest improved PFS and OS outcomes may be associated with newly approved treatments. In particular, the trends suggested that newer treatments may have a greater mPFS benefit vs. older treatments if used in earlier lines of treatment.Funding: GSK (HO-17-17767) [Display omitted] DisclosuresWillson:GlaxoSmithKline: Employment, Equity Ownership. Bruno:GlaxoSmithKline: Employment, Equity Ownership. Opalinska:GlaxoSmithKline: Employment, Equity Ownership. Nelson:GlaxoSmithKline: Equity Ownership. Stafkey-Mailey:Xcenda: Employment.

362. Species Distribution and Trends of Invasive Candidiasis in the United States, 2009–2015, Using a Large Electronic Medical Record Database

BackgroundWhile 50% of invasive candidiasis (IC) has historically been caused by C. albicans, the changing epidemiology and rise in drug-resistant Candida necessitates understanding the contribution of specific Candida species to IC. To date, species and site-specific trends in IC have not been reported on a large scale using US clinical data.MethodsUsing the Cerner Health Facts electronic health record (EHR) dataset, inpatient hospitalizations with any Candida spp. isolated from blood or a sterile site (SS) (abdominal or other) were identified from 2009 to 2015. Patient characteristics were described by species. Significant relationships (P ≤ 0.05) were assessed using chi-squared or exact binomial tests. Annual percent change in IC incidence by site and species were assessed via Poisson regression.ResultsOverall, 19,310 Candida isolates from 10,313 patients were identified. Of these, 46% of isolates were C. albicans, 22% C. glabrata, 14% C. parapsilosis, 7% C. tropicalis, and 11% other/unspeciated; no C. auris infections were identified. The overall incidence of IC was 99 cases/100,000 patients. Compared with C. albicans, isolation of other species was 35% more frequent from blood, and 43% and 30% less frequent from non-blood abdominal and non-abdominal SSs, respectively (Table 1). Total IC increased by 1% (95% CI = 0.2–2%) annually; while abdominal and SS IC significantly increased by 6% (4–8%) and 11% (9–13%) per year, respectively, candidemia decreased significantly by 4.5% (3–6%) annually. Among C. albicansinfections candidemia decreased by 6.5% (5–8%) annually, while abdominal (5%, 3–8%) and other SS infections (10%, 7–13%) increased (Figure 1). Candidemia incidence remained unchanged for the other species. SS infections increased for every species, and abdominal infections increased for all but C. parapsilosis (Figure 2).ConclusionIn this first large-scale study on trends in IC using US hospital EHR data, the species distribution of IC isolates varied between blood and non-blood SSs. The incidence of candidemia is decreasing, but not for potentially drug-resistant species such as C. glabrata, which continue to pose treatment challenges.This work was supported in part by the Division of Intramural Research, NIAID, NIH Disclosures All authors: No reported disclosures.

Validity of diagnostic codes and estimation of prevalence of diabetic foot ulcers using a large electronic medical record database.

To assess the validity of the diagnostic codes relating to diabetic foot ulcer (DFU) in the electronic medical records of a large integrated care provider and to assess the prevalence of DFU among its members. Data were obtained from the diabetes registry of Maccabi Healthcare Services (MHS), a 2.1-million-member sick-fund in Israel, which included 125665 patients in 2015. We randomly selected and reviewed ~400 patient files from each of the following categories during study period: (1) had a diagnostic code of DFU; (2) had a diagnostic code, or clinical condition suggestive of DFU including: leg-ulcer, amputation, DFU in quartiles proximate to 2015 or abnormality reported by nurse; (3) patients at high risk for DFU (age>35 and one of the following: peripheral artery disease, neuropathy, DFU during 2011-2014, eGFR<30mL/min/m2 or foot deformity). The patients' charts were reviewed by study physicians, and DFU was validated or refuted. Relying upon diagnostic codes entered by physicians, the positive predictive value (PPV) was 73.1% (95% CI 67.6-78.2), and the sensitivity was 48.2% (95% CI 45.8-50.7%). The PPV of the diagnostic codes listed by podiatrists were significantly lower, while that of codes listed by nurses was higher but with lower sensitivity. The estimated annual prevalence of DFU in the diabetes registry of MHS was 1.2% (95%CI 1.0-1.5%). Diagnostic codes alone cannot be used reliably to create a DFU registry. Nevertheless, the data collected provide an estimate of the prevalence of DFU among patients included in the MHS diabetes registry.

1512O - Automated survival prediction in metastatic cancer patients using high-dimensional electronic medical record data

Background: Oncologists use patients’ life expectancy to guide medical decisions and may benefit from a tool that provides accurate, unbiased assessments of prognosis. Existing prognostic models generally use only a few predictor variables. We used a large electronic medical record dataset to train a prognostic model for patients with metastatic cancer. Methods: The model was trained and tested using data from 12,588 patients treated for metastatic cancer in the Stanford Health Care system from 2008-2017. Data sources included provider note text, labs, vital signs, procedures, medication orders, and diagnosis codes. Patients were divided randomly into training and test sets (80%/20% split). A regularized Cox proportional hazards model with 4,126 predictor variables was fit to the training set and evaluated on the test set. A landmarking approach was used due to the multiple observations per patient, with t0 set to the time of metastatic cancer diagnosis. Performance was also evaluated using 399 palliative radiation courses in test set patients. An existing published model that uses performance status, primary tumor site, and treated site was used as a baseline [Chow, JCO 2008:20;26(36)]. Results: From the first visit after metastatic cancer diagnosis, median follow-up was 14.5 months and median overall survival was 20.9 months. Patients were seen for 384,402 daily visits. The prognostic model’s C-index for overall survival was 0.79 in the test set (averaged across landmark times). For palliative radiation courses, the C-index was 0.75 (95% CI 0.72-0.78), compared to 0.64 (95% CI 0.60-0.67) for an existing published model (p < 0.001).Table: 1512OPredicted vs actual survival for 2,518 test set patients at landmark time t0 (first visit after diagnosis of metastatic cancer)Predicted median survival in monthsActual median survival in months (95% CI)0-3 (n = 106)1.3 (0.9-2.0)3.1-6 (n = 172)3.7 (2.5-5.2)6.1-12 (n = 382)6.7 (5.6-7.9)>12 (n = 1858)35.7 (31.8-39.1) Open table in a new tab Conclusions: The model showed high predictive performance, which was significantly better than that of an existing model. Because it is fully automated, the model can be used to examine providers’ practice patterns and could deployed in a decision support tool to help improve quality of care. Legal entity responsible for the study: Michael Gensheimer. Funding: National Institutes of Health. Disclosure: M.F. Gensheimer: Research funding: Varian Medical Systems, Philips Healthcare. E. Cho: Employee of Genentech. D. Rubin: Research funding: Philips Healthcare. D.T. Chang: Research funding, honoraria: Varian Medical Systems; Stock ownership: ViewRay. All other authors have declared no conflicts of interest.

Machine learning: how much does it improve the prediction of unplanned hospital admissions?

IntroductionRisk prediction models can be used to inform decision-making in clinical settings. With large and detailed electronic medical record data, machine learning may improve predictions. The objective of this work is to determine the feasibility and accuracy of machine learning versus logistic regression to predict unplanned hospital admissions.&#x0D; Objectives and ApproachData from primary care electronic medical records for community-dwelling adults in Alberta, Canada available from the Canadian Primary Care Sentinel Surveillance Network will be linked to acute care administrative health data held by Alberta Health Services. Two regression methods (forward stepwise logistic, LASSO logistic) will be compared with three machine learning methods (classification tree, random forest, gradient boosted trees). Prior primary and acute care use will be used to predict three outcomes: ≥1 unplanned admission within 1 year, ≥1 unplanned admission within 90 days, and ≥1 unplanned admission within 1 year due to an ambulatory care sensitive condition.&#x0D; ResultsThe results of this work in progress will be presented at the conference. 41,142 patients will have their primary and acute care data linked. We anticipate that the machine learning methods will improve predictive performance but will be more challenging for clinicians and patients to understand, including why a given patient is predicted to be at higher risk. The primary comparison of machine learning and regression methods will be based on positive predictive values corresponding to the top 5% predicted risk threshold, and estimated via 10-fold cross-validation.&#x0D; Conclusion/ImplicationsThis project aims to help researchers decide which statistical methods to use for risk prediction models. When considering machine learning methods the best approach may be to try multiple methods, compare their predictive accuracy and interpretability, and then choose a final method.

Comparison of Characteristics of Benign Prostatic Hyperplasia (BPH) Patients Treated with Finasteride and Alpha Blocker Combination Therapy Versus Alpha Blocker Monotherapy in China: An Analysis of Electronic Medical Record Data.

Benign prostatic hyperplasia (BPH) is commonly treated with 5-alpha-reductase inhibitor/alpha blocker combination therapy or with alpha blocker monotherapy. However, in China, the characteristics of BPH patients receiving 5-alpha-reductase inhibitor/alpha blocker combination therapy or alpha blocker monotherapy remain largely unknown. Therefore, this study compared the characteristics of BPH patients receiving either the 5-alpha-reductase inhibitor finasteride in combination with an alpha blocker or an alpha blocker as monotherapy in clinical practice in China. Data were obtained from a large electronic medical record database from four tertiary hospitals in major cities in China (2009-2016). BPH patients aged ≥ 50years with ≥ 1 alpha blocker fill on/after the first BPH diagnosis were selected. Patients were further classified as receiving combination therapy (≥ 1 overlapping day of supply for finasteride and an alpha blocker) or alpha blocker monotherapy (did not receive any 5-alpha-reductase inhibitor). Patient characteristics, visit type (in- vs. outpatient) at treatment initiation, and comorbidities were evaluated during the 6-month baseline period and compared between the two groups using two sample t tests and chi-square tests/Fisher's exact tests. A total of 2666 and 2738 patients received combination therapy and monotherapy, respectively. The combination group was older (70.3 vs. 67.3years, p < 0.0001) and had more patients initiated in an inpatient setting (46.0% vs. 26.4%, p < 0.0001). Compared with the monotherapy group, the combination group had more comorbidities, such as hypertension (48.3% vs. 35.6%, p < 0.0001), cardiovascular disease (65.3% vs. 48.0%, p < 0.0001), and diabetes (21.1% vs. 15.7%, p < 0.0001), and a higher Charlson comorbidity index (0.9 vs. 0.7, p < 0.0001). Chinese BPH patients using finasteride/alpha blocker combination therapy were older and had a higher comorbidity burden than those using alpha blocker monotherapy. These findings provide Chinese healthcare decision-makers with a better understanding of the patient characteristics generally associated with BPH combination therapy vs. alpha blocker monotherapy. Merck Sharp and Dohme (China) Co., Ltd., Shanghai, China.

Development and validation of a heart failure with preserved ejection fraction cohort using electronic medical records

BackgroundHeart failure (HF) with preserved ejection fraction (HFpEF) comprises nearly half of prevalent HF, yet is challenging to curate in a large database of electronic medical records (EMR) since it requires both accurate HF diagnosis and left ventricular ejection fraction (EF) values to be consistently ≥50%.MethodsWe used the national Veterans Affairs EMR to curate a cohort of HFpEF patients from 2002 to 2014. EF values were extracted from clinical documents utilizing natural language processing and an iterative approach was used to refine the algorithm for verification of clinical HFpEF. The final algorithm utilized the following inclusion criteria: any International Classification of Diseases-9 (ICD-9) code of HF (428.xx); all recorded EF ≥50%; and either B-type natriuretic peptide (BNP) or aminoterminal pro-BNP (NT-proBNP) values recorded OR diuretic use within one month of diagnosis of HF. Validation of the algorithm was performed by 3 independent reviewers doing manual chart review of 100 HFpEF cases and 100 controls.ResultsWe established a HFpEF cohort of 80,248 patients (out of a total 1,155,376 patients with the ICD-9 diagnosis of HF). Mean age was 72 years; 96% were males and 12% were African-Americans. Validation analysis of the HFpEF algorithm had a sensitivity of 88%, specificity of 96%, positive predictive value of 96%, and a negative predictive value of 87% to identify HFpEF cases.ConclusionWe developed a sensitive, highly specific algorithm for detecting HFpEF in a large national database. This approach may be applicable to other large EMR databases to identify HFpEF patients.

Type 2 Diabetes Mellitus and Renal Impairment in Real-World with Large Outpatient Electronic Medical Records—Antihyperglycemic Medication Dose Adjustment

Objective: To estimate the level of inappropriate dosing of DPP-4 inhibitors, and to assess the level of emergency room visits, hypoglycemia, and mortality according to inappropriate dosing of DPP-4 inhibitors in type 2 diabetes (T2DM) patients with renal impairment(RI). Research Design and Methods: Retrospective database analysis using the National Health Insurance Service Database and the General Health Examination Database in Korea. Patients aged ≥ 40 years with evidence of T2DM (ICD-9-CM codes from E10.x-E14.x) January 1, 2009 to December 31, 2015. Renal impairment defined with estimated with the Modification of Diet in Renal Disease [MDRD] GFR equation. To estimate the level of inappropriate dosing of DPP-4 inhibitors in RI was according to daily prescribed dose. And The incidence of emergency room visits, hypoglycemia, and mortality were assessed using the hazard ratios (HRs) estimated with logistic regression model. Results: In our study, we analysis 161,760 patients with T2DM and RI. Among them DPP-4 inhibitors were prescribed to 44232 patients (27.3%). DPP-4 inhibitor prescribed 21.44% in GFR&amp;lt;30 and 32.28% in 30≤GFR&amp;lt;50. In observation period, percentages of inappropriate dosing of DPP-4 inhibitors in T2DM with RI was 31.9%. The age and sex adjusted HR with 95% confidence interval (CI) for appropriate dosing of DPP-4 inhibitors compared with inappropriate dosing of DPP-4 inhibitors was 0.871(0.842,0.9) for emergency room visits, 0.819(0.748,0.898) for sever hypoglycemia, and 0.775(0.717,0.839) for mortality. Conclusions: Renal impairment is common but often undetected in patients with T2DM. In Korea, one of three patients with T2DM and RI received inappropriate doses, and they were associated with high risk of emergency room visits and sever hypoglycemia, and higher mortality. Further analyses to understand the clinical and economic consequences of these findings are needed. Disclosure S. Hong: None. C. Park: None.