Electronic Medical Record Administrative Data Research Articles

The Predictive Value of Vital Signs for Morbidity in Pregnancy: Evaluating and Optimizing Maternal Early Warning Systems.

Vital sign scoring systems that alert providers of clinical deterioration prior to critical illness have been proposed as a means of reducing maternal risk. This study examined the predictive ability of established maternal early warning systems (MEWS)-as well as their component vital sign thresholds-for different types of maternal morbidity, to discern an optimal early warning system. This retrospective cohort study analyzed all patients admitted to the obstetric services of a four-hospital urban academic system in 2018. Three sets of published MEWS criteria were evaluated. Maternal morbidity was defined as a composite of hemorrhage, infection, acute cardiac disease, and acute respiratory disease ascertained from the electronic medical record data warehouse and administrative data. The test characteristics of each MEWS, as well as for heart rate, blood pressure, and oxygen saturation were compared. Of 14,597 obstetric admissions, 2,451 patients experienced the composite morbidity outcome (16.8%) including 980 cases of hemorrhage (6.7%), 1,337 of infection (9.2%), 362 of acute cardiac disease (2.5%), and 275 of acute respiratory disease (1.9%) (some patients had multiple types of morbidity). The sensitivities (15.3-64.8%), specificities (56.8-96.1%), and positive predictive values (22.3-44.5%) of the three MEWS criteria ranged widely for overall morbidity, as well as for each morbidity subcategory. Of patients with any morbidity, 28% met criteria for the most liberal vital sign combination, while only 2% met criteria for the most restrictive parameters, compared with 14 and 1% of patients without morbidity, respectively. Sensitivity for all combinations was low (maximum 28.2%), while specificity for all combinations was high, ranging from 86.1 to 99.3%. Though all MEWS criteria demonstrated poor sensitivity for maternal morbidity, permutations of the most abnormal vital signs have high specificity, suggesting that MEWS may be better implemented as a trigger tool for morbidity reduction strategies in the highest risk patients, rather than a general screen. · MEWS have poor sensitivity for maternal morbidity.. · MEWS can be optimized for high specificity using modified criteria.. · MEWS could be better used as a trigger tool..

Quantifying the Economic Impact of Delayed Multiple-Inhaler Triple Therapy Initiation in Patients with COPD: A Retrospective Cohort Study of Linked Electronic Medical Record and Hospital Administrative Data in England.

PurposeTo assess if early multiple-inhaler triple therapy (MITT) initiation in patients with chronic obstructive pulmonary disease (COPD) reduces subsequent healthcare resource utilization (HCRU), direct medical costs, and acute exacerbations of COPD (AECOPDs).Patients and MethodsThis retrospective, longitudinal cohort study used electronic health records and linked hospital administrative data in England. COPD patients with an AECOPD between July 2012 and May 2016 (index), and who subsequently started MITT within 180 days were eligible. Patients with an AECOPD 6 months prior to index were excluded. HCRU, direct healthcare costs, and AECOPDs were assessed in the following 24-month period for early (≤30 days) and delayed (31–180 days) MITT initiators.ResultsA total of 934 patients were included in the analysis and categorized as early (n=367, 39%) or delayed (n=567, 61%) MITT initiators. Mean patient age was 68.5 years and 53.2% were male. A significantly higher proportion of delayed MITT initiators required ≥1 outpatient appointment (all-cause) compared with early MITT initiators (87% vs 79%; p=0.0016). A significantly higher proportion of delayed MITT initiators required ≥1 COPD‑related inpatient stay versus early MITT initiators (47% vs 40%; p=0.0262). Over the 24-month follow-up, mean all-cause and COPD-related total healthcare costs were significantly higher in delayed MITT initiators compared with early MITT initiators (all‑cause: £11,348 vs £8126; p=0.0011; COPD-related: £7307 vs £4535; p=0.0009).ConclusionDelayed initiation of multiple-inhaler triple therapy was associated with higher all-cause and COPD-related costs, suggesting that earlier initiation of triple therapy in COPD patients may help reduce the economic burden on the healthcare system.

A hybrid model of acute unscheduled cancer care provided by a hospital-based acute care clinic and the emergency department: a descriptive study.

PurposePatients with cancer often experience medical events that require immediate evaluation. These evaluations typically occur in an emergency department (ED), but there is increasing interest in providing this care in other settings. We report on a novel care model whereby a nursing hotline is used to triage patients to the ED or to the North Carolina Cancer Hospital Infusion Center (NCCHIC).MethodsA retrospective study of adult patients with a neoplasm diagnosis seeking acute care at a large academic hospital pre- and post-initiation of the novel care model in January of 2016. Patients were identified by querying the electronic medical record and clinic administrative data during matched 6 month pre- and post-periods.ResultsDuring the pre-initiation period, 1346 patients visited the ED on 1651 occasions (76.1% admission rate). In the post-initiation period, 1434 patients visited the ED on 1797 occasions (81.5% admission rate), and 246 patients visited the NCCHIC on 322 occasions (68.9% admission rate). The emergency severity index (ESI) in the pre-initiation ED group was primarily ESI 2 (30.6%) and ESI 3 (65.4%). In the post-initiation ED group, the ESI was similar (32.6% ESI 2 and 64.2% ESI 3). In contrast, the NCCHIC predominantly treated lower acuity patients (65.8% calculated ESI of 4/5).ConclusionsThis model demonstrates a multidisciplinary partnership to providing acute unscheduled care for patients with cancer. In the early implementation phase of this model, approximately 15% of patients, generally of lower acuity, were seen in the NCCHIC.

Association of Roux-en-Y Gastric Bypass With Postoperative Health Care Use and Expenditures in Canada

Results of previous studies are mixed regarding the economic implications of a Roux-en-Y gastric bypass (RYGB). To assess the 5-year incremental health care use and expenditures after RYGB. This population-based cohort study conducted in Ontario, Canada, used a difference-in-differences approach to compare health care use and expenditures between patients who underwent a publicly funded RYGB from March 1, 2010, to March 31, 2013, and propensity score-matched control individuals who did not undergo a surgical bariatric procedure. The study period allowed for a minimum 60 months of follow-up because, at that time, the most recent date for which administrative data on health care and expenditures were available was March 31, 2018. Data sources included the Ontario Bariatric Registry linked to several Ontario health administrative databases and the Electronic Medical Record Administrative Data Linked Database. Health care use and expenditures data for 5 years before and 5 years after the index date (procedure date for RYGB group; random date for controls) were analyzed. Data analyses were performed March 12, 2019, to March 10, 2020. RYGB procedure. The primary outcome was total health care expenditures. The final propensity score-matched cohorts comprised 1587 individuals in the RYGB group (mean [SD] age, 47 [10.2] years) and 1587 controls (mean [SD] age, 47 [12.2] years); each group had 1228 women (77.4%) and a mean body mass index (calculated as weight in kilograms divided by height in meters squared) of 46. Mean total health care expenditures (2017 Canadian dollars) per patient in the RYGB group increased from CAD $15 594 (95% CI, CAD $14 743 to CAD $16 614) (US $12 008 [95% CI, US $11 353 to US $12 794]) in the 5 years before the procedure to CAD $30 389 (95% CI, CAD $28 789 to CAD $32 232) (US $23 401 [95% CI, US $22 169 to US $24 821]) over the 5 years after the procedure, a difference of CAD $14 795 (95% CI, CAD $13 172 to CAD $16 480) (US $11 393 [95% CI, US $10 143 to US $12 691]). For the control group, mean total health care expenditures per individual increased from CAD $16 109 (95% CI, CAD $14 727 to CAD $17 591) (US $12 405 [95% CI, US $11 341 to US $13 546]) 5 years before the index date to CAD $20 073 (95% CI, CAD $18 147 to CAD $22 169) (US $15 457 [95% CI, US $13 974 to US $17 071]) 5 years after the date, a difference of CAD $3964 (95% CI, CAD $2250 to CAD $5875) (US $3053 [95% CI, US $1733 to US $4524]). Overall, the difference-in-differences estimate of the net cost of RYGB was CAD $10 831 (95% CI, CAD $8252 to CAD $13 283) (US $8341 [95% CI, $6355 to $10 229]) over the 5-year period. This amount excluded the mean (SD) cost associated with the index date: CAD $6501 (CAD $1087) (US $5006 [US $837]) for the RYGB cohort and CAD $9 (CAD $72) (US $7 [US $55]) for the controls. The cost differential was primarily associated with increased hospitalizations in the first months immediately after RYGB. Expenditures leveled off in year 3 after the index date; differences in total expenditures between the RYGB and control cohorts were not statistically significantly different in years 4 and 5. Health care expenditures in the 3 years after publicly funded RYGB were higher in patients who underwent the procedure than in control individuals, but the costs were similar thereafter. This finding suggests the need to decrease hospital and emergency department readmissions after surgical bariatric procedures because such use is associated with increased spending.

Patient- and Physician-Level Factors Associated With Adherence to C-CHANGE Recommendations in Primary Care Settings in Ontario

BackgroundWe previously found large variation among family physicians in adherence to the Canadian Cardiovascular Harmonization of National Guidelines Endeavour (C-CHANGE). We assessed the role of patient- and physician-level factors in the variation in adherence to recommendations for managing cardiovascular disease risk factors. MethodsWe conducted a retrospective study using multilevel logistic regression analyses with the Electronic Medical Record Administrative data Linked Database (EMRALD) housed at ICES in Ontario. Five quality indicators based on C-CHANGE guidelines were modelled. Effects of clustering and between-group variation, patient-level (sociodemographics, comorbidities) and physician-level characteristics (demographic and practice information) were assessed to determine odds ratios of receiving C-CHANGE recommended care. ResultsIn all, 324 Ontario physicians practicing in 41 clinics who provided care to 227,999 adult patients were studied. We found significant variation in quality indicators, with 15% to 39% of the total variation attributable to nonpatient factors. The largest variation was in performing 2-hour plasma glucose testing in prediabetic patients. Patient-level factors most frequently associated with recommendation adherence included sex, age, and multi-comorbidities. Women were more likely than men to have their body mass index measured, and their blood pressure controlled, but less likely to receive antiplatelet medications and liver-enzyme testing if overweight or obese. ConclusionsThe majority of variations in adherence were attributable to patient attributes, but a substantial proportion of unexplained variation was due to differences among physicians and clinics. This finding may signal suboptimal processes or structures and warrant further investigation to improve the quality of primary care management of cardiovascular disease in Ontario.

Validity of early-onset dementia diagnoses in VA electronic medical record administrative data

Objective To determine the validity of diagnoses indicative of early-onset dementia (EOD) obtained from an algorithm using administrative data, we examined Veterans Health Administration (VHA) electronic medical records (EMRs). Method A previously used method of identifying cases of dementia using administrative data was applied to a random sample of 176 cases of Post-9/11 deployed veterans under 65 years of age. Retrospective, cross-sectional examination of EMRs was conducted, using a combination of administrative data, chart abstraction, and review/consensus by board-certified neuropsychologists. Results Approximately 73% of EOD diagnoses identified using existing algorithms were identified as false positives in the overall sample. This increased to approximately 76% among those with mental health conditions and approximately 85% among those with mild traumatic brain injury (TBI; i.e. concussion). Factors related to improved diagnostic accuracy included more severe TBI, diagnosing clinician type, presence of neuroimaging data, absence of a comorbid mental health condition diagnosis, and older age at time of diagnosis. Conclusions A previously used algorithm for detecting dementia using VHA administrative data was not supported for use in the younger adult samples and resulted in an unacceptably high number of false positives. Based on these findings, there is concern for possible misclassification in population studies using similar algorithms to identify rates of EOD among veterans. Further, we provide suggestions to develop an enhanced algorithm for more accurate dementia surveillance among younger populations.

SUN-154 Identifying Type 1 Diabetes in Electronic Medical Records and Administrative Health Data in Ontario: A Validation Study

Background There has been growing interest in using administrative and electronic medical record (EMR) data to identify type 1 diabetes (T1D) for clinical and health services research. Our aim was to create an algorithm identifying T1D using EMR data alone and/or in combination with administrative data in Ontario, Canada. Methods The Electronic Medical Record Administrative data Linked Database (EMRALD) was used to create a reference standard for diabetes type as determined by chart abstraction by an endocrinologist. EMRALD contains data from over 350 family physicians using Practice Solutions® EMR. We included patients receiving care from participating physicians who were diagnosed with diabetes prior to Sept 30, 2015. A random 25% sample of patients with diabetes (identified using a validated algorithm) were selected for the reference standard. Patients who were not prescribed insulin or who were prescribed insulin and an oral hypoglycemic agent other than metformin were automatically classified as not having T1D without chart abstraction. Remaining patients underwent chart abstraction to determine diabetes type. 50 charts were randomly re-abstracted for inter- and intra-rater reliability. Characteristics of patients by diabetes type were reviewed to select variables for algorithm inclusion. Preliminary Results There were 23,310 patients with diabetes as of September 30, 2015. The reference standard consisted of 5828 patients: 5305 were classified as not T1D without chart abstraction and 523 underwent chart abstraction. Inter- and intra-rater reliability were high (Cohen’s kappa 0.82 and 0.96, respectively). After chart abstraction,14 patients were classified as not having diabetes, 233 patients were classified as T1D (45.8%), 261 as not T1D (51.3%), and 15 as possible T1D (2.9%). The prevalence of T1D was 4.0%. Compared to 5581 patients without definite T1D, patients with T1D were younger (median 40 years [IQR 30-53] v. 66[57-75]), had a younger age of onset (15 years [10-25] v. 47[40-55]), and a lower body mass index (BMI) (26±5 v. 32±8). Age at diagnosis and BMI were documented in 349(71.8%) of abstracted charts and 1386(23.8%), respectively. Renal insufficiency (eGFR <30ml/min, dialysis, or renal transplant) was present in 13(5.6%) and 136(2.4%) patients with and without T1D, respectively. Metformin use was uncommon in patients with T1D compared to those without T1D [15(6.4%) v. 4223(75.7%)]. Bolus insulin was recorded by the primary care physician for 115(49.4%) patients with T1D and 490 (8.8%) patients without T1D. For patients with T1D, 65(27.9%) used an insulin pump and 41(21.4%) had a history of diabetic ketoacidosis. Conclusions Age, age at diagnosis, BMI, medications, and insulin pump use differed between subjects with and without T1D. Future work includes development of algorithms incorporating these variables and variables from other data sources to identify T1D.

Comparing prescribing and dispensing databases to study antibiotic use: a validation study of the Electronic Medical Record Administrative data Linked Database (EMRALD).

Monitoring and studying community antibiotic use is a critical component in combating rising antimicrobial resistance. To validate an electronic medical record dataset containing antibiotic prescriptions and to quantify some important differences between prescribing and dispensing databases. We evaluated antibiotics prescribed and dispensed to patients ≥65 years of age during 2011-15. We compared the EMRALD prescribing database with the validated Ontario Drug Benefit (ODB) dispensing database. Using ODB as the gold standard and limiting to EMRALD physicians, we calculated sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) with 95% CIs. We also compared the relative change in antibiotic use prescribed by all physicians to this population over time between the databases using Poisson regression models. In this population, 74% of all antibiotics dispensed were from non-EMRALD physicians. Trends in use were discordant over time. When we limited ODB to EMRALD prescribers only to assess the validity of EMRALD data, we observed good sensitivity and excellent specificity for correctly identifying antibiotics at 85% (95% CI 84%-85%) and 98% (95% CI 98%-98%), respectively. The PPV was 78% (95% CI 78%-78%) and the NPV was 99% (95% CI 99%-99%). All performance measures were higher among the highest prescribing physicians. We demonstrated EMRALD is well suited for studying antibiotic prescribing by EMRALD physicians. However, due to the frequency with which patients receive antibiotic prescriptions from their non-primary care physicians, we caution against the use of non-population-based prescribing databases to infer antibiotic use rates or trends over time.

Temporal trends in severe obesity prevalence in children and youth from primary care electronic medical records in Ontario: a repeated cross-sectional study.

There are no current estimates of severe obesity in Canadian children. The objectives of this study were to determine the prevalence of severe obesity in children aged 18 years or less in Ontario and to determine temporal trends from 2004 to 2015. This was a repeated cross-sectional study using height/length and weight of children aged 18 years or less from the Electronic Medical Record Administrative data Linked Database (EMRALD), a database of primary care electronic medical records in Ontario. We calculated body mass index (for age and sex) z-scores (zBMI). Two years of data (2014 and 2015) were used to determine the period prevalence of severe obesity. We used multivariable linear regression generalized estimating equations to estimate the association of calendar year and mean zBMI. In total, 55 233 children were included. The prevalence of severe obesity (zBMI > 3) increased with increasing age: it was 0.9% (95% confidence interval [CI] 0.7% to 1.0%) among children less than 5 years of age, 2.7% (95% CI 2.3% to 3.1%) among 5- to 9-year-olds, 2.9% (95% CI 2.4% to 3.3%) among 10- to 14-year-olds and 3.7% (95% CI 3.1% to 4.3%) among those aged 15-18. Boys aged 5-9 years had a significantly higher prevalence of severe obesity than their female counterparts (3.5% [95% CI 2.9% to 4.2%] v. 1.7% [95% CI 1.3% to 2.2%]). From 2004 to 2015, the mean zBMI decreased by 0.015 (95% CI -0.018 to -0.012) units per year, with the overall prevalence of severe obesity in all ages highest in 2005 (3%) and a decrease to 2% in 2015. The prevalence of severe obesity among children and adolescents in Ontario is consistent with that in other developed countries with the exception of the United States. There is evidence of plateauing of estimates and a small decrease in zBMI over time. Further understanding of the impact of prevention efforts on these estimates is an important next step.

Measuring Cardiovascular Quality in Primary Care Using Canadian Cardiovascular Harmonization of National Guidelines Endeavour and Electronic Medical Record Data in Ontario

BackgroundThis project uses electronic medical record (EMR) data to assess performance by family physicians (FPs) in the screening for, diagnosis, and management of cardiovascular disease (CVD) and risk factors against national harmonized guidelines by the Canadian Cardiovascular Harmonization of National Guidelines Endeavour (C-CHANGE). MethodsA retrospective cohort study using the Electronic Medical Record Administrative Data Linked Database (EMRALD) was conducted. A set of quality indicators (QIs) were developed on the basis of the 2014 C-CHANGE guidelines. Twenty-three readily measurable QIs were used to measure performance in the screening for and management of CVD, and to identify gaps in performance. ResultsOur study population consisted of 324 Ontario FPs and 284,959 patients. We assessed 23 of the 74 recommendations. There was variance in rates of adherence to QIs related to screening rates for CVD. Highest adherence to C-CHANGE guidelines was related to laboratory testing for patients with hypertension and prescription of antihypertensive therapies (≥ 91.4%). Lowest adherence to the guidelines was seen in administration of oral glucose tolerance tests for assessing prediabetic patients (4.4%). ConclusionsFP EMR data can be used to measure adherence to one-third of the C-CHANGE recommendations. There are varying levels of adherence among the measurable C-CHANGE recommendations, and there is room for improvement in quality of primary care management of CVD in Ontario. There is potential to use EMR data to assess changes to CVD management in FP practice using guidelines if recommendations are quantifiable and measurable.

Use of physician billing claims to identify infections in children.

While medical records have detailed information, they are limited in reach to the availability and accessibility of those records. On the other hand, administrative data while limited in scope, have a much further reach in coverage of an entire population. However, few studies have validated the use of administrative data for identifying infections in pediatric populations. Pediatric patients from Ontario, Canada aged <18 years were randomly sampled from the Electronic Medical Record Administrative data Linked Database (EMRALD). Using physician diagnoses from the electronic medical record (EMR) as the reference standard, we determined the criterion validity of physician billing claims in administrative data for identifying infectious disease syndromes from 2012 to 2014. Diagnosis codes were assessed by infection category (respiratory, skin and soft tissue, gastrointestinal, urinary tract and otitis externa) and for all infections combined. Sensitivity analyses assessed the performance if patients had more than one reason to visit the physician. We analysed 2,139 patients and found 33.3% of all visits were for an infection, and respiratory infections accounted for 67.6% of the infections. When we combined all infection categories, sensitivity was 0.74 (95% CI 0.70–0.77), specificity was 0.95 (95% CI 0.93–0.96), positive predictive value (PPV) was 0.87 (95% CI 0.84–0.90), and negative predictive value (NPV) was 0.88 (95% CI 0.86–0.89). For respiratory infections, sensitivity was 0.77 (95% CI 0.73–0.81), specificity was 0.96 (95% CI 0.95–0.97), PPV was 0.85 (95% CI 0.81–0.88), and NPV was 0.94 (95% CI 0.92–0.95). Similar performance was observed for skin and soft tissue, gastrointestinal, urinary tract, and otitis externa infections, but with lower sensitivity. Performance measures were highest when the patient visited the physician with only one health complaint. We found when using linked EMR data as the reference standard, administrative billing codes are reasonably accurate in identifying infections in a pediatric population.

Identify Patients with Congestive Heart Failure through Analyzing Free-Text Clinical Notes

IntroductionA number of challenges exist in analyzing unstructured free text data in electronic medical records (EMRs). EMR text are difficult to represent and model due to their high dimensionality, heterogeneity, sparsity, incompleteness, random errors and the presence of noise.&#x0D; Objectives and ApproachStandard Natural Language Processing (NLP) tools make errors when applied to clinical notes due to physician use of unconventional language, involving polysemy, abbreviations, ambiguity, misspelling, variations, and negation.&#x0D; This paper presents a novel NLP framework, “Clinical Learning On Natural Expression” (CLONE), to automatically learn from a large primary care EMR database, analyzing free text clinical notes from primary care practices. CLONE’s predictive clinical models using text mining and neural network approach to extract features to identify patterns. To demonstrate effectiveness, we evaluate CLONE’s ability in a case study to identify patients with a specific chronic condition: congestive heart failure (CHF).&#x0D; ResultsA random selected sample of 7500 patients from Electronic Medical Record Administrative data Linked Database (EMRALD) is used. In this dataset, each patient’s medical chart includes a reference standard, manually reviewed by medical practitioners. Prevalence of CHF is approximately 2%. The low prevalence leads to another challenging problem in machine learning: imbalanced datasets. After pre-processing, we build deep learning models to represent and extract important medical information from free text to identify CHF patients through analyzing patient charts. We evaluated the effectiveness of CLONE by comparing the predicted labels with the standard references on a holdout test dataset. Comparing it with a number of alternative algorithms, we improve the overall accuracy to over 90% on a test dataset.&#x0D; Conclusion/ImplicationsAs the role of NLP in EMR data expands, the CLONE natural language processing framework can lead to substantial reduction in manual processing, while improving predictive accuracy.

Use of linked electronic health records to evaluate cardiovascular risk prediction models in Ontario, Canada

IntroductionElectronic health records (EHR) contain individual-level clinical information not found in traditional administrative databases. As part of the CANHEART-Strategy for Patient Oriented Research (SPOR) initiative, we created a linked EHR-administrative data cohort that enables us to measure the Framingham and ACC/AHA Pooled Cohort cardiovascular risk prediction scores in Ontario, Canada.&#x0D; Objectives and ApproachAn EHR primary care cohort was created using the Electronic Medical Record Administrative data Linked Database (EMRALD) database, which contains the blood pressure and lipid values, weight and height measures, prescriptions and smoking status of up to 350,000 patients in Ontario, Canada. We enriched the lipid information through linkage to the Ontario Laboratory Information System, which is a repository of 90%+ of all lipid tests in Ontario. Individual-level information on co-morbidities, hospitalizations and mortality attributed to cardiovascular causes (e.g. myocardial infarction, stroke, cardiovascular mortality) were obtained through linkage to provincial health administrative and vital statistics databases using CANHEART methodology (www.canheart.ca).&#x0D; ResultsPatients were entered into the cohort between 2008 and 2014 if they had measurements for blood pressure and lipids (total cholesterol and high-density lipoprotein) taken within a year of each other during this accrual window. The earliest such group of values was chosen and determined the individual’s index date. Age, sex, smoking, diabetes and anti-hypertensive treatment status were extracted from EHR or administrative data to calculate the two scores. Patients were excluded if not aged 40-75 on the index date or if they had a history of cardiovascular disease. A cohort of 84,628 Ontario residents (mean age 55.0 years) had the elements required to calculate both scores. Follow-up for outcome events were done through record linkage to the end of 2014, with a mean follow-up of 3.62 years.&#x0D; Conclusion/ImplicationsThe creation of this cohort will allow for the validation of the Framingham and AHA/ACC Pooled Cohort equations in the diverse Ontario population. It would also enable the possible development of a new ‘made-in-Canada’ cardiovascular risk prediction model.

Using family physician Electronic Medical Record data to measure the pathways of cancer care

&#x0D; IntroductionGaps in care have been identified along the disease pathway for specific cancers. However, no real-world data exists to identify wait times along these cancer pathways. Secondary use of family physician (FP) electronic medical record data (EMR) can augment existing health administrative data in measuring steps in the care pathways.&#x0D; Objectives and ApproachWe used FP EMR data to identify care pathways for lung cancer and breast cancer patients from the description of symptoms, to the initiation of investigations, referrals to specialty care and the receipt of specific treatments (surgery, chemotherapy, radiation treatment). Data from the Electronic Medical Record Administrative data Linked Database (EMRALD) held at the Institute for Clinical Evaluative Sciences (ICES) was used to identify a cohort of lung cancer and breast cancer patients. Data abstractors examined the FP EMR notes to identify pre-diagnostic symptoms, pre-diagnostic radiological test, biopsy results, oncology and surgical specialist referrals and post-diagnostic surgical and oncological consultations.&#x0D; ResultsTo date, abstractors have reviewed the FP EMR notes for 300 lung cancer patient and 1200 breast cancer patients. Abstractors identified an index date where there was documentation of the first abnormal test result and/or a FP progress note documenting a “suspicious” or “concerning” sign or symptom. For both lung cancer and breast cancer patients, a pre-diagnostic index date was identified in 88.5% of FP EMR notes. For lung cancer patients 66.7% based were based on abnormal chest x-rays and for breast cancer patients 81.1% were based on abnormal mammograms. Pre-diagnostic symptoms were identified in 62.1% of FP EMR notes and 81.6% had post-diagnostic consultation notes. Wait times from the index date to seeing an oncological specialist were less than four weeks for all patients.&#x0D; Conclusion/ImplicationsWe are able to use information from FP EMRs linked to health administrative data to identify pre-diagnostic care received by patients prior to their cancer diagnosis. This information can be used to identify care gaps and measure wait times in receiving cancer care from a patient’s perspective.

Using Biomedical Text as Data and Representation Learning for Identifying Patients with an Osteoarthritis Phenotype in the Electronic Medical Record

IntroductionElectronic medical records (EMRs) are increasingly used in health services research. Accurate/efficient identification of a target population with a specific disease phenotype is a necessary precursor to studying the health of these individuals.&#x0D; Objectives and ApproachWe explored the use of biomedical text as inputs to supervised phenotype identification algorithms. We employed a two-stage classification approach to map the discrete, sparse high-dimensional biomedical text data to a dense low dimensional vector space using methods from unsupervised machine learning. Next we used these learned vectors as inputs to supervised machine learning algorithms for phenotype identification.&#x0D; We were able to demonstrate the applicability of the approach to identifying patients with an osteoarthritis (OA) phenotype using primary care data from the Electronic Medical Record Administrative data Linked Database (EMRALD) held at ICES.&#x0D; ResultsEMRALD contains approximately 20Gb of biomedical text data on approximately 500,000 patients. The unit of analysis for this study is the patient. We were interested in identifying OA patients using solely text data as features.&#x0D; Labelled outcome information wass available from a random sample of 7,500 patients. We divided patients into training (N=6000), validation (N=750) and test (N=750) cohorts. We learned low dimensional representations of the input text data on the entire EMRALD corpus (N=500,000). We used learned numeric vectors as inputs to supervised machine learning models for OA classification (N=6,000 training set patients).&#x0D; We compared models in terms of accuracy, sensitivity, specificity, PPV and NPV. The best learned models achieved approximately 90\% sensitivity and 80\% specificity. Classification accuracy varied as a function of learned inputs.&#x0D; Conclusion/ImplicationsWe developed an approach to phenotype identification using solely biomedical text as an input. Preliminary results suggest our two-stage ML approach has improved operating characteristics compared to existing clinically derived decision rules for OA classification. Future work will explore the generalizability of this methodology to other disease phenotypes.

Rotavirus vaccine coverage and factors associated with uptake using linked data: Ontario, Canada.

BackgroundIn August 2011, Ontario, Canada introduced a rotavirus immunization program using Rotarix™ vaccine. No assessments of rotavirus vaccine coverage have been previously conducted in Ontario.MethodsWe assessed vaccine coverage (series initiation and completion) and factors associated with uptake using the Electronic Medical Record Administrative data Linked Database (EMRALD), a collection of family physician electronic medical records (EMR) linked to health administrative data. Series initiation (1 dose) and series completion (2 doses) before and after the program’s introduction were calculated. To identify factors associated with series initiation and completion, adjusted odds ratios (aOR) and 95% confidence intervals (95%CI) were calculated using logistic regression.ResultsA total of 12,525 children were included. Series completion increased each year of the program (73%, 79% and 84%, respectively). Factors associated with series initiation included high continuity of care (aOR = 2.15; 95%CI, 1.61–2.87), maternal influenza vaccination (aOR = 1.55; 95%CI,1.24–1.93), maternal immmigration to Canada in the last five years (aOR = 1.47; 95% CI, 1.05–2.04), and having no siblings (aOR = 1.62; 95%CI,1.30–2.03). Relative to the first program year, infants were more likely to initiate the series in the second year (aOR = 1.71; 95% CI 1.39–2.10) and third year (aOR = 2.02; 95% CI 1.56–2.61) of the program. Infants receiving care from physicians with large practices were less likely to initiate the series (aOR 0.91; 95%CI, 0.88–0.94, per 100 patients rostered) and less likely to complete the series (aOR 0.94; 95%CI, 0.91–0.97, per 100 patients rostered). Additional associations were identified for series completion.ConclusionsFamily physician delivery achieved moderately high coverage in the program’s first three years. This assessment demonstrates the usefulness of EMR data for evaluating vaccine coverage. Important insights into factors associated with initiation or completion (i.e. high continuity of care, smaller roster sizes, rural practice location) suggest areas for research and potential program supports.

Veracity in big data: How good is good enough.

Veracity, one of the five V's used to describe big data, has received attention when it comes to using electronic medical record data for research purposes. In this perspective article, we discuss the idea of data veracity and associated concepts as it relates to the use of electronic medical record data and administrative data in research. We discuss the idea that electronic medical record data are "good enough" for clinical practice and, as such, are "good enough" for certain applications. We then propose three primary issues to attend to when establishing data veracity: data provenance, cross validation, and context.

Infant feeding practices within a large electronic medical record database

BackgroundThe emerging adoption of the electronic medical record (EMR) in primary care enables clinicians and researchers to efficiently examine epidemiological trends in child health, including infant feeding practices.MethodsWe completed a population-based retrospective cohort study of 8815 singleton infants born at term in Ontario, Canada, April 2002 to March 2013. Newborn records were linked to the Electronic Medical Record Administrative data Linked Database (EMRALD™), which uses patient-level information from participating family practice EMRs across Ontario. We assessed exclusive breastfeeding patterns using an automated electronic search algorithm, with manual review of EMRs when the latter was not possible. We examined the rate of breastfeeding at visits corresponding to 2, 4 and 6 months of age, as well as sociodemographic factors associated with exclusive breastfeeding.ResultsOf the 8815 newborns, 1044 (11.8%) lacked breastfeeding information in their EMR. Rates of exclusive breastfeeding were 39.5% at 2 months, 32.4% at 4 months and 25.1% at 6 months. At age 6 months, exclusive breastfeeding rates were highest among mothers aged ≥40 vs. < 20 years (rate ratio [RR] 2.45, 95% confidence interval [CI] 1.62–3.68), urban vs. rural residence (RR 1.35, 95% CI 1.22–1.50), and highest vs. lowest income quintile (RR 1.18, 95% CI 1.02–1.36). Overall, immigrants had similar rates of exclusive breastfeeding as non-immigrants; yet, by age 6 months, among those residing in the lowest income quintile, immigrants were more likely to exclusively breastfeed than their non-immigrant counterparts (RR 1.43, 95% CI 1.12–1.83).ConclusionsWe efficiently determined rates and factors associated with exclusive breastfeeding using data from a large EMR database.

Completeness and accuracy of anthropometric measurements in electronic medical records for children attending primary care

BackgroundElectronic medical records (EMRs) from primary care may be a feasible source of height and weight data. However, the use of EMRs in research has been impeded by lack of...

Primary Care Screening and Comorbidity Management in Rheumatoid Arthritis in Ontario, Canada.

Quality measurement for rheumatoid arthritis (RA) patients has largely focused on care provided by rheumatologists. Our aim was to develop and assess quality measures related to the screening and management of comorbidity in RA patients in primary care. We used the primary care Electronic Medical Record Administrative data Linked Database in Ontario, Canada. We harmonized Canadian general population and RA clinical recommendations to develop and assess screening, process, and outcome measures. For each RA patient, 10 non-RA patients were matched by age and sex. Stratified analyses were performed, comparing patients with RA to those without RA, to assess the performance of quality measures. We compared 1,405 RA patients to 14,050 matched non-RA patients (72.8% female; mean age 62.5 years). Compared to non-RA patients, RA patients more frequently had influenza (44.9% versus 40.0%) and pneumococcal (40.4% versus 34.1%) vaccinations and bone mineral density testing (67.4% versus 58.1%). Herpes zoster vaccinations were less frequent among RA patients (13.8% versus 19.5%), as was screening for cervical cancer (58.6% versus 64.0%). No significant differences were observed between RA and non-RA patients in screenings for breast (70.7% versus 73.8%) or colorectal (31.7% versus 34.5%) cancers. Only a quarter of RA patients had a comprehensive cardiovascular risk assessment. No definitive differences were detected in the management of patients who had co-occurring cardiovascular disease or diabetes mellitus. For both RA and non-RA patients, compliance with Canadian recommendations for preventive medical services and screening for comorbid conditions in primary care was less than optimal. This indicates key targets for improvement.