Primary Surveillance Research Articles

Influenza epidemics observed in primary care from 1984 to 2017 in France: A decrease in epidemic size over time.

BackgroundEpidemiological analysis of past influenza epidemics remains essential to understand the evolution of the disease and optimize control and prevention strategies. Here, we aimed to use data collected by a primary care surveillance system over the last three decades to study trends in influenza epidemics and describe epidemic profiles according to circulating influenza viruses.MethodsInfluenza‐like illness (ILI) weekly incidences were estimated using cases reported by general practitioners participating in the French Sentinelles network, between 1984 and 2017. Influenza epidemics were detected by applying a periodic regression to this time series. Epidemic (co‐)dominant influenza virus (sub)types were determined using French virology data.ResultsDuring the study period, 297 607 ILI cases were reported allowing the detection of 33 influenza epidemics. On average, seasonal epidemics lasted 9 weeks and affected 4.1% of the population (95% CI 3.5; 4.7). Mean age of cases was 29 years. Epidemic size decreased over time by ‐66 cases per 100 000 population per season on average (95% CI −132; −0.2, P value = 0.049) and epidemic height decreased by ‐15 cases per 100 000 (95% CI −28; −2, P value = 0.022). Epidemic duration appeared stable over time. Epidemics were mostly dominated by A(H3N2) (n = 17, 52%), associated with larger epidemic size, higher epidemic peak and older age of cases.ConclusionsThe declining trend in influenza epidemic size and height over the last 33 years might be related to several factors like increased vaccine coverage, hygiene improvements or changing in influenza viruses. However, further researches are needed to assess the impact of potential contributing factors to adapt influenza plans.

Towards a Knowledge-Based Recommender System for Linking Electronic Patient Records With Continuing Medical Education Information at the Point of Care

Given the limits of human memory, clinicians have trouble recalling therapeutic recommendations, even when the clinician previously judged that the information relevant for the care of a specific patient. To tackle this problem, we present a knowledge-based recommender system prototype that links the electronic patient records to clinical information, previously delivered to the target physician and judged to be potentially beneficial. We developed this prototype within the context of RxTx, a Canadian continuing medical education program. We apply a constraint-based recommendation strategy as follows: (1) clinical experts (taggers) map a set of therapeutic recommendations (called Highlights) to a requirement statement built from the standard clinical codes and supplementary demographic information, when applicable; (2) a matching system identifies patient-Highlights recommendation pairs through requirement satisfaction; and (3) given a patient record being examined, the recommended Highlights can be retrieved online at the point of care. We tested this prototype using electronic medical records from the Canadian Primary Care Sentinel Surveillance Network and 87 therapeutic Highlights from the RxTx collection, evaluating the system’s performance against a gold standard consisting of a two-expert consolidated patient-Highlight matching set for 150 patient records. The requirements-based recommendation system exhibits very high precision (mode: 1.0, 89% of the time; average precision: 0.95) and moderate recall (mode 1.0, 48.7% of the time; average recall: 0.61). The near-perfect precision minimizes the possibility of generating alert fatigue in physicians using the system. We note that more than half of the false negative results from the information being available in the text of the electronic medical records, but unavailable as a clinical code. The near-perfect precision over the tested patient set suggests that the system has the potential to deliver high-quality recommendations of clinical information at the point of care while being easily integrated within a continuing medical education program and the clinician’s workflow.

Persistence of leprosy in household social networks: overlapping cases and vulnerability in endemic regions in Brazil

The study sought to analyze the magnitude of occurrence and the sociodemographic, economic and clinical profiles of leprosy associated with household social networks (HSN), with disease overlap in cities from the states of Bahia, Piauí and Rondônia, Brazil, from 2001 to 2014. This is a cross-sectional study using primary and secondary data regarding new cases of leprosy notified to the Brazilian Information System for Notifiable Diseases (SINAN, in Portuguese) residing in the cities. We applied a standardized instrument to the new cases and reviewed data from charts and from SINAN. Of a total of 1,032 (29.6%) assessed cases, 538 (52.1%) had more than one case in their HSN. There were larger frequencies of female sex (292; 54.3%), age between 41 and 60 years (240; 44.6%), primary education (272; 50.6%), income lower than the minimum wage (265; 49.3%) and living with 5 or more people (265; 49.3%). The overlap of cases in the HSN was associated in the multivariate analysis with residing in cities in the state of Rondônia (PR = 1.23; 95%CI: 1.07-1.43; p = 0.003), as well as living with 3 to 4 people in the same household (PR = 1.66; 95%CI: 1.11-2.49; p = 0.014) and having leprosy reaction (PR = 1.31; 95%CI: 0.99-1.70; p = 0.050). Case repetition within the same HSN is a frequent event in the situations we studied. Its occurrence must be considered as a sentinel indicator of greater epidemiological severity in primary health care surveillance. We highlight the vulnerability of affected families.

Post-operative thyroid US evaluation

Thyroid cancer is the most common endocrine malignancy. The increased incidence of differentiated thyroid carcinoma (DTC) in the past few decades is likely due to more intensive imaging and the detection of early stage disease. Local recurrence can be found in the thyroid bed or the residual thyroid tissue and regional recurrence in the lymph nodes of the central or lateral compartments of the neck. Although recurrences are relatively common and may occur years after initial treatment, mortality rates of DTC remain very low. The primary surveillance paradigm now used for most treated DTC patients at low risk for disease recurrence consists of serial serum thyroglobulin measurements and focused cervical US. Neck ultrasonography (US) is recommended for the assessment of all patients with thyroid carcinoma after initial therapy, since even low-risk patients with undetectable stimulated thyroglobulin (Tg) may present cervical metastases. Cervical lymph node metastases have been reported to occur in 12–81% of patients with papillary thyroid cancer (PTC), and in a smaller proportion of patients with other histotypes (i.e., follicular thyroid cancer and Hu¨rthle cell carcinoma) Besides, recurrences, there are other differential diagnostic for thyroid bed nodules, as enlarged parathyroid and lymph nodes. Specific US findings of recurrent tumor in the neck include cystic changes, microcalcifications, and clustering of rounded lymph nodes. The color Doppler US can be useful helping to distinguish between malignant and benign nodules (both in the thyroid bed and lymph nodes). The flow analysis usually revealing peripheral or mixed hypervascularization in malignant cases. In the presence of “suspicious” lymph nodes upon US, fine-needle aspiration cytology and measurement of Tg in the needle lavage fluid are useful and complementary exams for the definition of the etiology, with the combination of the two methods showing elevated sensitivity and 100% specificity.

Prediction of fatty liver disease using machine learning algorithms

Background and objectiveFatty liver disease (FLD) is a common clinical complication; it is associated with high morbidity and mortality. However, an early prediction of FLD patients provides an opportunity to make an appropriate strategy for prevention, early diagnosis and treatment. We aimed to develop a machine learning model to predict FLD that could assist physicians in classifying high-risk patients and make a novel diagnosis, prevent and manage FLD. MethodsWe included all patients who had an initial fatty liver screening at the New Taipei City Hospital between 1st and 31st December 2009. Classification models such as random forest (RF), Naïve Bayes (NB), artificial neural networks (ANN), and logistic regression (LR) were developed to predict FLD. The area under the receiver operating characteristic curve (ROC) was used to evaluate performances among the four models. ResultsA total of 577 patients were included in this study; of those 377 patients had fatty liver. The area under the receiver operating characteristic (AUROC) of RF, NB, ANN, and LR with 10 fold-cross validation was 0.925, 0.888, 0.895, and 0.854 respectively. Additionally, The accuracy of RF, NB, ANN, and LR 87.48, 82.65, 81.85, and 76.96%. ConclusionIn this study, we developed and compared the four classification models to predict fatty liver disease accurately. However, the random forest model showed higher performance than other classification models. Implementation of a random forest model in the clinical setting could help physicians to stratify fatty liver patients for primary prevention, surveillance, early treatment, and management.

Combining primary care surveillance and a meta-analysis to estimate the incidence of the clinical manifestations of Lyme borreliosis in Belgium, 2015–2017

Lyme borreliosis (LB) is an important tick-borne disease which can cause a broad range of symptoms mainly affecting the skin, the nervous system and the joints. This study aims to estimate the incidence of the different clinical manifestations of LB in Belgium. The incidence of erythema migrans (EM) was estimated through the network of sentinel general practices at 97.6/100,000 inhabitants (uncertainty interval [UI] 82.0–113.0) for the period 2015–2017. This result was used to estimate the incidence of other LB manifestations based on their proportional distribution (ratios) to EM reported in the neighboring countries of Belgium. To estimate these ratios, we performed a systematic review of studies published between February 1, 2008 and January 31, 2018 and pooled the results using a random effects meta-analysis. Six studies were retained in the systematic review, and the meta-analysis estimated the occurrence ratios for Lyme neuroborreliosis/EM, Lyme arthritis/EM and other manifestations/EM at 0.024 (95% confidence interval [CI] 0.016–0.037), 0.022 (95% CI 0.020–0.024) and 0.014 (95% CI 0.012–0.016) respectively. Applying these ratios to the EM incidence in Belgium resulted in an incidence estimation of 2.4/100,000 inhabitants (95% UI 1.5–3.7) for Lyme neuroborreliosis, 2.1/100,000 (95% UI 1.7–2.6) for Lyme arthritis and 1.4/100,000 (95% UI 1.1–1.7) for other less frequent manifestations. Some of these LB manifestations, other than EM, are more severe, hence these estimates are essential to assess the health burden and economic cost of LB which would be highly relevant for patients, healthcare providers and policymakers. As both over- and underestimation of different clinical LB manifestations remain possible due to characteristics of the primary surveillance systems and the disease itself, future studies to validate these estimates would be of great value.

Mucopolysaccharidosis type II detection by Naïve Bayes Classifier: An example of patient classification for a rare disease using electronic medical records from the Canadian Primary Care Sentinel Surveillance Network.

Identifying patients with rare diseases associated with common symptoms is challenging. Hunter syndrome, or Mucopolysaccharidosis type II is a progressive rare disease caused by a deficiency in the activity of the lysosomal enzyme, iduronate 2-sulphatase. It is inherited in an X-linked manner resulting in males being significantly affected. Expression in females varies with the majority being unaffected although symptoms may emerge over time. We developed a Naïve Bayes classification (NBC) algorithm utilizing the clinical diagnosis and symptoms of patients contained within their de-identified and unstructured electronic medical records (EMR) extracted by the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). To do so, we created a training dataset using published results in the scientific literature and from all MPS II symptoms and applied the training dataset and its independent features to compute the conditional posterior probabilities of having MPS II disease as a categorical dependent variable for 506497 male patients. The classifier identified 125 patients with the highest likelihood for having the disease and 18 features were selected to be necessary for forecasting. Next, a Recursive Backward Feature Elimination algorithm was employed, for optimal input features of the NBC model, using a k-fold Cross-Validation with 3 replicates. The accuracy of the final model was estimated by the Validation Set Approach technique and the bootstrap resampling. We also investigated that whether the NBC is as accurate as three other Bayesian networks. The Naïve Bayes Classifier appears to be an efficient algorithm in assisting physicians with the diagnosis of Hunter syndrome allowing optimal patient management.

Estimating severity of influenza epidemics from severe acute respiratory infections (SARI) in intensive care units

BackgroundWhile influenza-like-illness (ILI) surveillance is well-organized at primary care level in Europe, few data are available on more severe cases. With retrospective data from intensive care units (ICU) we aim to fill this current knowledge gap. Using multiple parameters proposed by the World Health Organization we estimate the burden of severe acute respiratory infections (SARI) in the ICU and how this varies between influenza epidemics.MethodsWe analyzed weekly ICU admissions in the Netherlands (2007–2016) from the National Intensive Care Evaluation (NICE) quality registry (100% coverage of adult ICUs in 2016; population size 14 million) to calculate SARI incidence, SARI peak levels, ICU SARI mortality, SARI mean Acute Physiology and Chronic Health Evaluation (APACHE) IV score, and the ICU SARI/ILI ratio. These parameters were calculated both yearly and per separate influenza epidemic (defined epidemic weeks). A SARI syndrome was defined as admission diagnosis being any of six pneumonia or pulmonary sepsis codes in the APACHE IV prognostic model. Influenza epidemic periods were retrieved from primary care sentinel influenza surveillance data.ResultsAnnually, an average of 13% of medical admissions to adult ICUs were for a SARI but varied widely between weeks (minimum 5% to maximum 25% per week). Admissions for bacterial pneumonia (59%) and pulmonary sepsis (25%) contributed most to ICU SARI. Between the eight different influenza epidemics under study, the value of each of the severity parameters varied. Per parameter the minimum and maximum of those eight values were as follows: ICU SARI incidence 558–2400 cumulated admissions nationwide, rate 0.40–1.71/10,000 inhabitants; average APACHE score 71–78; ICU SARI mortality 13–20%; ICU SARI/ILI ratio 8–17 cases per 1000 expected medically attended ILI in primary care); peak-incidence 101–188 ICU SARI admissions in highest-incidence week, rate 0.07–0.13/10,000 population).ConclusionsIn the ICU there is great variation between the yearly influenza epidemic periods in terms of different influenza severity parameters. The parameters also complement each other by reflecting different aspects of severity. Prospective syndromic ICU SARI surveillance, as proposed by the World Health Organization, thereby would provide insight into the severity of ongoing influenza epidemics, which differ from season to season.

Development of Optical Fiber Connected Passive Primary Surveillance Radar

Development of Optical Fiber Connected Passive Primary Surveillance Radar

Impact of the Diabetes Canada Guideline Dissemination Strategy on the Prescription of Vascular Protective Medications: A Retrospective Cohort Study, 2010-2015.

The 2013 Diabetes Canada guidelines launched targeted dissemination tools and a simple assessment for vascular protection. We aimed to 1) examine changes associated with the launch of the 2013 guidelines and additional dissemination efforts in the rates of vascular protective medications prescribed in primary care for older patients with diabetes and 2) examine differences in the rates of prescriptions of vascular protective medications by patient and provider characteristics. The study population included patients (≥40 years of age) from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) with type 2 diabetes and at least one clinic visit from April 2010 to December 2015. An interrupted time series analysis was used to assess the proportion of eligible patients prescribed a statin, ACE inhibitor (ACEI)/angiotensin receptor blocker (ARB), or antiplatelet prescription in each quarter. Proton pump inhibitor (PPI) prescriptions were the reference control. A dynamic cohort was used where participants were enrolled each quarter using a prespecified set of conditions (range 25,985-70,693 per quarter). There were no significant changes in statin (P = 0.43), ACEI/ARB (P = 0.42), antiplatelet (P = 0.39), or PPI (P = 0.16) prescriptions at baseline (guideline intervention). After guideline publication, there was a significant change in slope for statin (-0.52% per quarter, SE 0.15, P < 0.05), ACEI/ARB (-0.38% per quarter, SE 0.13, P < 0.05), and reference PPI (-0.18% per quarter, SE 0.05, P < 0.05) prescriptions. There was a decrease in prescribing trends over time that was not specific to vascular protective medications. More effective knowledge translation strategies are needed to improve vascular protection in diabetes in order for patients to receive the most effective interventions.

Primary care-based surveillance to estimate the proportion of rotavirus gastroenteritis among Latvian children below 5 years of age with acute gastroenteritis

ABSTRACT Background: Rotavirus (RV) is worldwide an important cause of acute gastroenteritis (AGE) in infants and young children. There is no specific treatment for AGE caused by RV (RVGE) but since 2006 two safe and effective vaccines have been available. RV vaccination was included in the national immunization program (NIP) of Latvia in 2015 with full reimbursement, and within the first year a coverage of 87% was achieved. This surveillance study was carried out to investigate the proportion of RVGE among AGE episodes in Latvia up to the inclusion of RV vaccination in the NIP to provide a basis for future assessments of the impact of RV vaccination. Methods: Prospective, one-year observational study of children younger than 5 years presenting with AGE in the primary care setting. At first primary care contact, a stool sample was collected and tested for RV using a rapid, visual immunochromatographic kit. The parents monitored their child’s symptoms over 2 weeks after the first contact and the investigator recorded these observations during a follow-up phone call. The proportion of RVGE among the AGE cases was estimated and the severity of each AGE case was assessed based on the recorded symptoms using the 20-point Vesikari scale. The seasonality of RVGE was also investigated. Results: Fifty-two primary care investigators collected data on 606 evaluable children with AGE. The proportion of RVGE was 38.1%. Severe AGE was experienced by 40.7% of the RV-positive and 19.5% of the RV-negative patients. The rate of hospitalization was 9.1% for the RV-positive and 4.8% for the RV-negative with no difference in the mean duration of hospital stays. AGE and RVGE both occurred all year round but with a clearly marked peak only for RVGE, from March to May. Conclusion: This study underlines that RV is an important cause of AGE in children under 5 years old in Latvia and that the burden of disease of RVGE in primary care was substantial before inclusion of RV vaccination in the NIP. Trial registration: NCT01733849

Herpes Zoster Burden in Canadian Provinces: A Narrative Review and Comparison with Quebec Provincial Data.

Background The main aim of this review was to assess incidence rates and trends of medically attended and death cases of herpes zoster in Canada. Methods The search was conducted in five databases (PubMed, Cochrane, Embase, PsycNET, and Web of Science). Data on herpes zoster-related consultations and hospitalisations and deaths were also extracted from three Quebec provincial administrative databases (RAMQ, MED-ECHO, and ISQ). Results The electronic search yielded 587 publications. Seventeen publications satisfied inclusion criteria. These publications reported data from eleven studies. Ten studies used provincial databases, and one study used the Canadian Primary Care Sentinel Surveillance Network electronic database. Seven studies evaluated overall rates of medically attended cases (consultations and hospitalisations). Four of these studies reported an increase in rates of medically attended cases during the study period; one study reported stable rates, and two studies reported only an average rate. The rates varied from 316 to 450/100,000 p.y. The Quebec analysis shows similar rates with a slight decreasing trend (from 369 to 350/100,000 p.y.). Incidence rates of consultations were reported separately in three studies. Two studies reported an increase in rates (from 258 to 348/100,000 p.y. and from 324 to 366/100,000 p.y.), and the third study reported a decrease (from 525 to 479/100,000 p.y.). Hospitalization rates were reported separately in two studies, both reporting a decrease (from 12 to 8 cases/100,000 p.y. and from 9 to 4 cases/100,000 p.y.). Quebec data also showed a decrease, from 9 to 6 cases/100,000 p.y. One study reported herpes zoster-related deaths. In this study, the reported death rate was 0.7/1,000,000 p.y. in the overall population and 5.5/1,000,000 p.y. in those aged ≥65 years. Quebec analysis showed a death rate of 1.2/1,000,000 p.y. in the overall population and 8.6/1,000,000 p.y. in those aged ≥65 years. Conclusions The results of the reviewed studies and our analysis of Quebec provincial data indicate important variations in the reported overall incidence rates of medically attended herpes zoster cases in Canada. The trends in time are heterogeneous in studies in which hospitalisations and medical consultations were pooled together. We observed a decrease in hospitalization rates and a slight increase in consultation rates in studies reporting hospitalisations and consultations separately. These results consolidate the understanding of the herpes zoster burden in Canada and might be used as a tool in decision-making regarding future preventive interventions.

Linking primary care EMR data and administrative data in Alberta, Canada: experiences, challenges, and potential solutions

IntroductionAdministrative data are commonly used for a variety of secondary purposes. Although they lack clinical detail and risk factor information, linkage to primary care electronic medical records (EMR) could fill this gap. Primary care EMRs are a relatively new data source available in Alberta and thus, EMR-administrative linkages are novel.&#x0D; Objectives and ApproachTo describe the process undertaken for linking de-identified primary care EMR data from two regional Alberta networks of the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) with administrative data (hospital admissions, emergency department visits, pharmacy information) from Alberta Health Services Analytics, specifically as it relates to a study on patients with complex, chronic diseases. As this linkage process is new in Alberta, we will describe the challenges encountered and possible solutions to inform future data linkage for research studies.&#x0D; ResultsLinkage steps: 1) approval from research ethics board and individual CPCSSN providers as data custodians; 2) notify Privacy Commissioner on behalf of custodian; 3) send linking key (CPCSSN patient ID, EMR ID) from regional database to Analytics; 4) send linking files (patient personal health number [PHN], EMR ID) from custodian’s EMR system to Analytics; 5) match unique EMR ID from linking key and clinic linking files; 6) PHN from clinic linking file mapped to administrative data; 7) data de-identified before transferring to secure repository; administrative data matched to EMR data using CPCSSN ID.&#x0D; Challenges: obtaining individual provider consent for each study; sampling bias; delays/issues generating clinic linkage file; mismatch between patients in clinic \&amp; regional linking files.&#x0D; Current and potential solutions will be discussed during the presentation.&#x0D; Conclusion/ImplicationsAs primary care EMR and administrative data become more routinely linked and accepted, the process will become more efficient and streamlined. These data will contribute to a better understanding of patients and their care in Alberta.

Design and validation of a universal influenza virus enrichment probe set and its utility in deep sequence analysis of primary cloacal swab surveillance samples of wild birds

Influenza virus infections in humans and animals are major public health concerns. In the current study, a set of universal influenza enrichment probes was developed to increase the sensitivity of sequence-based virus detection and characterization for all influenza viruses. This universal influenza enrichment probe set contains 46,953 120nt RNA biotin-labeled probes designed based on all available influenza viral sequences and it can be used to enrich for influenza sequences without prior knowledge of type or subtype. Marked enrichment was demonstrated in influenza A/H1N1, influenza B, and H1-to-H16 hemagglutinin plasmids spiked into human DNA and in cultured influenza A/H2N1 virus. Furthermore, enrichment effects and mixed influenza A virus infections were revealed in wild bird cloacal swab samples. Therefore, this universal influenza virus enrichment probe system can capture and enrich influenza viral sequences selectively and effectively in different samples, especially ones with degraded RNA or containing low amount of influenza RNA.

Developing and implementing linked electronic medical record and administrative data in primary care practice for diabetes in Alberta

IntroductionUse of administrative health data and primary care electronic medical record data are both ubiquitous in Alberta, but linkage between them at patient level and implementation of the linked data into primary care practice are rare. This demonstration project sought to achieve this for a sample of patients with diabetes.&#x0D; Objectives and ApproachAcademic family physicians in the Department of Family Medicine at the University of Calgary who participate in the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) identified diabetes–related variables, either in their EMRs or in administrative data, that they wished to obtain in a linked dataset. Secure data linkage was obtained through Alberta Health Services (the provincial health authority) following transmission of patient mapping files direct from the clinics. The de-identified, linked, patient data was then transferred to CPCSSN-Alberta data managers for processing and displayed to users through an interactive Diabetes Dashboard.&#x0D; Results2598 patients with diabetes were identified using a validated CPCSSN case definition from 47 family physicians in three clinics. CPCSSN EMR data included primary care encounters, date of diagnosis, deprivation index, BMI, blood pressure, comorbidity, diabetes medications prescribed, risk factors, etc. Administrative data included laboratory results (HbA1c, fasting blood glucose, cholesterol, triglycerides, creatinine), medication dispensed, emergency room visits, inpatient admissions and costs. Integrated, interactive provider reports were created and sent to participating physicians. The reports presented the information about diabetes patients at individual provider level, bench-marked at clinic, primary care network and provincial levels. Follow-up with providers led to further dashboard development . We propose to scale up implementation of the integrated diabetes database and dashboard to include all 23,000 CPCSSN-identified diabetes patients in Alberta.&#x0D; Conclusion/ImplicationsIntegration of EMR and administrative data and its application to clinical care, panel management, and quality improvement in primary care, as well as to surveillance and research, was feasible and acceptable to the family physicians participating in this project.

Can Linked Electronic Medical Record and Administrative Data Help Us Identify Those Living With Frailty?

IntroductionFrailty is a combination of factors that increase vulnerability to functional decline, dependence and/or death. Frailty cannot easily be defined by comorbidities or medical treatment alone. Accurate detection of frailty in practice and at a population level is needed. This may be achieved using a combination of data sources.&#x0D; Objectives and ApproachWe construct algorithms that can identify frailty using electronic medical record (EMR) and administrative data. We linked EMR data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) nodes and the administrative (e.g. billings, hospitalizations) from Population Data BC and the Manitoba Health Policy Centre. Frailty was defined as individuals 65+ who were receiving home services, had specific chronic conditions, received specific diagnoses, and/or had specific lab or other clinical indicators. We describe sociodemographic characteristics, risk factors, prescribed medications, use and costs of healthcare for those identified as frail.&#x0D; ResultsPeople were identified as frail in 2014 and all analysis was completed with 2015 data. Among those who were &gt; 65 years, who had a record in both EMR and administrative data, 5\%-8\% (n=191 of 3,553, BC; n=2,396 of 29,382, MB) were identified as frail. There was a higher likelihood of being frail with increasing age and being a woman. In BC, those identified as frail have higher contacts with primary care (n=20 vs. n=10) and more days in hospital (n=7.4 vs. n=2.0) compared to those who are not frail. Twenty two percent of those identified as frail in 2014 died in 2015, compared to a mortality rate of 2\% among those who are not frail.&#x0D; Conclusion/ImplicationsIdentifying and reporting on those who are frail in primary care as well as in communities could enable targeted communications with patients and families and community based resources in order to improve patient care, patients’ and caregivers’ quality of life and better use of the healthcare system.

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.

Developing a Primary Care EMR-based Frailty Definition using Machine Learning

IntroductionFrailty is a geriatric syndrome that is predictive of heightened vulnerability for disability, hospitalization, and mortality. Annually an estimated 250,000 frail Canadians die, and this estimate is expected to double in the next 40 years, as Canadians grow older. Currently there is no single accepted clinical definition of frailty.&#x0D; Objectives and ApproachThe objective of this study was to develop an operational definition of frailty using machine learning that can be applied to a primary care electronic medical record (EMR) database.&#x0D; The Canadian Primary Care Sentinel Surveillance Network (CPCSSN) is a pan-Canadian network of primary care practices that collect de-identified patient information (such as encounter diagnoses, health conditions, and laboratory data) from EMRs.&#x0D; 780 patients from CPCSSN have were randomly selected and assessed by physicians using the Rockwood Clinical Frailty Scale (as frail or not frail), and their clinical characteristics from CPCSSN used to develop the definition using machine-learning.&#x0D; ResultsA total of 8,044 clinical features were extracted from these tables: billing, problem list, encounter diagnosis, labs, medications and referrals. A chi-squared automatic interaction detector (CHAID) approach was selected as the best approach. The bootstrapping process used a cost matrix that prioritized high sensitivity and positive predictive value. 10-fold cross validation was used for validity measures. Key features factored into the algorithm included: diagnosis of dementia (ICD-9 code 290), medications furosemide and vitamins, and use of key word “obstruction” within the billing table. The validation measures with 95% confidence intervals are as follows: sensitivity of 28% (95% CI: 21% to 36%), specificity of 94% (95% CI: 93% to 96%), positive predictive value of 53% (95% CI: 42% to 64%), negative predictive value of 86% (95% CI: 83% to 88%).&#x0D; Conclusion/ImplicationsNo other primary care specific frailty screening tools have sufficient validity. These results suggest heterogeneous diseases require clearly defined features and potentially more sophisticated algorithms to account for heterogeneity. Further research utilizing continuous features and continuous frailty scores may be more suitable in the creation of a case detection algorithm.

Association rule mining to identify potential under-coding of conditions in the problem list in primary care electronic medical records

IntroductionThe problem list of a patient’s primary care electronic medical record (EMR) generally reflects their important medical conditions. We will use association rule mining to assess between-provider and between-clinic variation in the coding of select conditions in the EMR problem list, in order to identify possible under-coding outliers.&#x0D; Objectives and ApproachEMR data from participating clinics in the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) will be used, with a focus on three commonly-occurring conditions (hypertension, diabetes, and depression). Association rule mining will be used to develop association rules between these conditions and other clinical information available in the EMR, such as other diagnoses in the problem list, billing codes, medications, and laboratory results (e.g., a rule of “diabetic medication→diabetes” indicates that patients prescribed a diabetic medication are likely to have diabetes in the problem list). Under-coding outliers at the provider and clinic levels will be identified by comparing rule enforcement.&#x0D; ResultsResults from this work in progress will be presented at the conference. An estimated 270 clinics, 1340 providers, and 1.8 million patients will be included from the CPCSSN database. Rule ‘confidence’ will be used to identify outliers; the confidence of a rule X→Y is the proportion of individuals with X who also have Y (Pr(Y|X)). For example, we may find that on average 80% of patients prescribed a diabetic medication will also have a diagnosis of diabetes in the problem list (average confidence of 80%), but an outlier clinic may have a confidence of 40%; this low rule confidence may indicate under-coding of diabetes in the problem list. Confounding by patient demographics (e.g., age, sex, urban/rural) will be assessed and adjusted for, if necessary.&#x0D; Conclusion/ImplicationsThis work examines a novel method to identify potential under-coding in the EMR problem list. Providers/clinics could use this information to update patients’ problem list or inform quality improvement interventions. Researchers using primary care EMR data need to be aware of potential under-coding and take steps to mitigate the effects.

Principles and operational model for governing Diabetes Action Canada's data repository for patient-oriented research

Introduction Diabetes Action Canada is developing a data repository and registry of potential research participants to support research, QI, and service to improve diabetes care. Central to the repository are pseudonymised linkable electronic medical records (EMRs) from family practices that are participating in the Canadian Primary Care Sentinel Surveillance Network (CPCSSN).&#x0D; Objectives and Approach We sought to develop an information governance process that would engender the trust of patients and the health care professionals (HCPs) that their EMR data were being managed responsibly in the best interests of patients living with diabetes. Following an extensive literature review, we developed a principles-based governance framework and operational model, with a strong focus on patient participation in the governance process. We recruited patients through our pre-existing patient advisory circles and physicians through our partners in CPCSSN. In January 2018, we held a training workshop for Research Governing Committee (RGC) members.&#x0D; Results We identified eight values-based principles to guide our governance process: transparency; accountability; following the rule of law; integrity of purpose, science and ethics; participation and inclusiveness; impartiality and independence; effectiveness; efficiency and responsiveness; and reflexivity and continuous quality improvement of process. Patients represent 50% of RGC members and HCPs 20%. Patient members provide their perspectives on: goals and outcomes of the research; the benefits and burdens among people living with diabetes; and the communication preferences of patients around recruitment. HCPs provide a deep understanding of the settings and systems in which care is provided to ensure contextual integrity of the research. Two researchers and one person with bioethics expertise provide technical and ethics perspectives on data requests.&#x0D; Conclusion/Implications Governance must go beyond legal compliance to ensure a ’social licence’ for the use of the data. In part, we address this through our guiding principles, our emphasis on patient and healthcare provider perspectives, and focus on research that is scientifically sound, ethically robust and in the public interest.