Validation of case-ascertainment algorithms using health administrative data to identify people who inject drugs in Ontario, Canada

OBJECTIVE

Health administrative data can be used to improve the health of people who inject drugs by informing public health surveillance and program planning, monitoring, and evaluation. However, methodological gaps in the use of these data persist due to challenges in accurately identifying injection drug use at the population level. In this study, we validated case-ascertainment algorithms for identifying people who inject drugs using health administrative data in Ontario, Canada.

STUDY DESIGN AND SETTING

Data from cohorts of people with recent (past 12 month) injection drug use, including those participating in community-based research studies or seeking drug treatment were linked to health administrative data in Ontario from 1992-2020. We assessed the validity of algorithms to identify injection drug use over varying lookback periods (i.e., all years of data [1992 onwards] or within the past 1-5 years), including inpatient and outpatient physician billing claims for drug use, emergency department visits or hospitalizations for drug use or injection-related infections, and opioid agonist treatment (OAT).

RESULTS

Algorithms were validated using data from 15,241 people with recent IDU (918 in community cohorts, 14,323 seeking drug treatment). An algorithm consisting of ≥1 physician visit, emergency department visit or hospitalization for drug use, or OAT record could effectively identify IDU history (91.6% sensitivity, 94.2% specificity) and recent IDU (using 3 years lookback: 80.4% sensitivity, 99% specificity) among community cohorts. Algorithms were generally more sensitive among people who inject drugs seeking drug treatment.

CONCLUSION

Validated algorithms using health administrative data performed well in identifying people who inject drugs. Despite high sensitivity and specificity, the positive predictive value of these algorithms will vary depending on the underlying prevalence of injection drug use in the population in which they are applied.

Predictors and disparities in neonatal and under 5 mortality in rural Pakistan: cross sectional analysis

Background

Adopted in 2015, the sustainable development goals (SDGs) have set specific targets (SDG 3.2) for countries to reduce their neonatal mortality rate (NMR) to below 12 deaths per 1000 live births and under 5 mortality rate (U5MR) to below 25 deaths per 1000 live births by 2030. For Pakistan to achieve these targets, there is a need to measure these rates and understand the predictors of child mortality at sub-national level. Launched in 2016, the Umeed-e-Nau (UeN) or New Hope project is based on scaling up proven and effective Maternal and Newborn Child Health (MNCH) interventions in 8 of the highest burden districts of the country, using existing public sector platforms in Pakistan at both the community and facility level. The primary aim of the project is to reduce perinatal mortality in these districts by 20% from baseline.

Methods

We report overall neonatal and post neonatal mortality rates for the two years preceding the UeN baseline household survey. Rates were calculated using the synthetic cohort probability method and predictors of neonatal and post neonatal mortality examined using Cox regression. To investigate spatial variations in the mortality rates, we calculated Moran's I at the district level using predicted probabilities of mortality. Finally, we create district level maps of predicted under 5 child mortality using a stochastic partial differentiation approach.

Findings

A total of 26,258 children contributed to the analysis of mortality with 838 deaths in the neonatal period and 2236 under-5 deaths during the observation period from March 1, 2015 to March 17, 2017. Overall, we estimated the NMR to be 29.2 per 1000 live births (95% CI: 26.9-31.4) and the U5MR to be 86.1 per 1000 live births (95% CI: 85.5-86.8). We found evidence of within-district geospatial clustering of under 5 mortality (P < 0.0001) and that social factors (poverty, illiteracy, multiparity), poor coverage of community health workers and distance from health facilities were strongly associated with child mortality.

Interpretation

Important factors associated with neonatal and post-neonatal mortality in our study population included maternal education, parity, household size and gender. Additionally, antenatal care coverage (at least 4 visits) was specifically associated with neonatal mortality only, whereas, LHW coverage and distance to health facility were strongly associated with post-neonatal mortality. These findings emphasise the need for comprehensive, multisectoral strategies to be implemented for future maternal and child health programs and outreach services in rural areas.

Funding

The study was funded by an unrestricted grant from the Bill & Melinda Gates Foundation to the Aga Khan University (Grant OPP 1148892).

Assessing transcriptomic reidentification risks using discriminative sequence models

Gene expression data provide molecular insights into the functional impact of genetic variation, for example, through expression quantitative trait loci (eQTLs). With an improving understanding of the association between genotypes and gene expression comes a greater concern that gene expression profiles could be matched to genotype profiles of the same individuals in another data set, known as a linking attack. Prior works show such a risk could analyze only a fraction of eQTLs that is independent owing to restrictive model assumptions, leaving the full extent of this risk incompletely understood. To address this challenge, we introduce the discriminative sequence model (DSM), a novel probabilistic framework for predicting a sequence of genotypes based on gene expression data. By modeling the joint distribution over all known eQTLs in a genomic region, DSM improves the power of linking attacks with necessary calibration for linkage disequilibrium and redundant predictive signals. We show greater linking accuracy of DSM compared with existing approaches across a range of attack scenarios and data sets including up to 22,288 individuals, suggesting that DSM helps uncover a substantial additional risk overlooked by previous studies. Our work provides a unified framework for assessing the privacy risks of sharing diverse omics data sets beyond transcriptomics.

Evaluating the impact of a virtual outpatient care program in preventing hospitalizations, emergency department visits and mortality for patients with COVID-19: a matched cohort study

OBJECTIVES

We sought to evaluate the impact of virtual care in preventing unnecessary healthcare visits for SARSCoV-2 patients.

METHODS

We conducted a retrospective matched cohort study, evaluating the COVID-19 Expansion to Outpatients (COVIDEO) program involving virtual assessments for all positive patients in the Sunnybrook assessment center from January 2020 to June 2021, followed by risk-stratified routine follow-up, couriering of oxygen saturation devices, and 24 hour/day direct-to-physician pager for urgent questions. We linked COVIDEO data to province-wide datasets, matching each eligible COVIDEO patient to ≤10 other Ontario SARS-CoV-2 patients on age, sex, neighborhood, and date. The primary outcome was emergency department (ED) visit, hospitalization or death within 30 days. Multivariable regression accounted for comorbidities, vaccination and pre-pandemic healthcare utilization.

RESULTS

Among 6,508 eligible COVIDEO patients, 4,763 (73.1%) were matched to ≥1 non-COVIDEO patient. COVIDEO care was protective against the primary composite outcome (adjusted odds ratio (aOR) 0.91, 95%CI 0.82-1.02), with a reduction in ED visits (7.8% vs 9.6%; aOR 0.79, 95%CI 0.70-0.89), but increase in hospitalizations (3.8% vs 2.7%, aOR 1.37, 95%CI 1.14-1.63) reflecting more direct-to-ward admissions (1.3% vs 0.2%, p < 0.0001). Results were similar when matched comparators were limited to patients who had not received virtual care elsewhere with a decrease in ED visits (7.8 vs. 8.6%, aOR 0.86, 95%CI 0.75-0.99) and an increase in hospitalizations (3.7 vs 2.4%, aOR 1.45, 95%CI 1.17-1.80).

CONCLUSIONS

An intensive remote care program can prevent unnecessary ED visits and facilitate direct-to-ward hospitalizations and thereby mitigate the impact of COVID-19 on the healthcare system.

Antimicrobial resistance and mortality following E. coli bacteremia

BACKGROUND

Global estimates suggest millions of deaths annually are associated with antimicrobial resistance (AMR) but these are generated from scarce data on the relative risk of death attributable to drug-resistant versus drug-sensitive infections.

METHODS

We examined all episodes of E. coli bloodstream infection in Ontario, Canada between 2017 and 2020, and measured 90 day mortality among those with resistant versus sensitive isolates for each of 8 commonly used antibiotic classes and a category of difficult to treat resistance (DTTR). We used multivariable logistic regression to calculate an adjusted odds of mortality associated with AMR, after accounting for patient demographics, comorbidities, and prior healthcare exposure.

FINDINGS

Among 14,548 eligible episodes of E. coli bloodstream infection, resistance was most common to aminopenicillins (46.8%), followed by first generation cephalosporins (38.8%), fluoroquinolones (26.5%), sulfonamides (24.1%), third generation cephalosporins (13.8%), aminoglycosides (11.7%), beta-lactam-beta-lactamase-inhibitors (9.1%) and carbapenems (0.2%). Only 18 (0.1%) episodes exhibited DTTR. For each antibiotic class, the unadjusted odds of mortality (OR) were higher among resistant isolates, but after accounting for patient characteristics the adjusted odds (aOR) of mortality were attenuated: aminopenicillins (OR 1.22, 95% CI 1.12-1.33; aOR 1.09, 95% CI 0.99-1.20), first generation cephalosporins (OR 1.24, 95% CI 1.14-1.35; aOR 1.07, 95% CI 0.97-1.18), third generation cephalosporins (OR 1.64, 95% CI 1.47-1.82; aOR 1.29, 95% CI 1.15-1.46), beta-lactam-beta-lactamase-inhibitors (OR 1.69, 95% CI 1.52-1.89, aOR 1.28, 95% CI 1.13-1.45), carbapenems (OR 3.11, 95% CI 1.52-6.34; aOR 2.06, 95% CI 0.91-4.66), sulfonamides (OR 1.19, 95% CI 1.07-1.31, aOR 1.06, 95% CI 0.95-1.18), fluoroquinolones (OR 1.49, 95% CI 1.36-1.64, aOR 1.16, 95% CI 1.05-1.29), aminoglycosides (OR 1.43, 95% CI 1.27-1.62; aOR 1.27, 95% CI 1.11-1.46), and DTTR (OR 3.71, 95% CI 1.46-9.41; aOR 2.58, 95% CI 0.87-7.66).

INTERPRETATION

AMR is associated with substantial increased mortality among patients with E. coli bloodstream infection, particularly for resistance to classes commonly used as empiric treatment. Surveillance for AMR-associated mortality should incorporate adjustment for patient characteristics and prior healthcare utilization.

FUNDING

This work was supported by a project grant from CIHR (grant number 159503). This study was also supported by ICES, which is funded by an annual grant from Ontario Ministry of Health and Long-Term Care (MOHLTC).

Pool size measurements improve precision of flux estimates but increase sensitivity to unmodeled reactions outside the core network in isotopically nonstationary metabolic flux analysis (INST-MFA)

Metabolic flux analysis (MFA) involves model-based estimation of metabolic reaction rates (i.e., fluxes) and, in some cases, metabolite content (i.e., pool sizes) from experimental measurements. Applying MFA to biological data helps determine the fate of substrates and the activity of specific pathways within metabolic networks. However, reliably estimating fluxes by using simplified "core" models to predict the dynamics of larger metabolic networks remains a challenge. One point of uncertainty relates to the advantages and potential pitfalls of including pool size measurements as experimental inputs for isotopically nonstationary MFA (INST-MFA). Here, we directly assessed the role of pool sizes using various core models and simulated datasets. To investigate the effects of pool size measurements on INST-MFA, we assessed the accuracy and precision of flux estimates obtained using different subsets of data (e.g., with or without pool size measurements) and simple network models that either matched or differed from the true network. The inclusion of pool size measurements provided incremental improvements to the precision of the flux estimates. However, adding pool size measurements increased the sensitivity of the flux solution to unmodeled reactions outside the core network. These results were confirmed using a large E. coli model that is representative of realistic metabolic networks examined in MFA studies. Our findings indicate that accurate flux estimates can be obtained in the absence of pool size measurements, even when using core models that lack full network coverage. Addition of pool size measurements to INST-MFA datasets may reveal the activity of non-core reactions that influence the labeling dynamics and therefore necessitate network expansion in order to reconcile all available data to the model. Our findings also emphasize the key role that goodness-of-fit testing plays in assessing the quality of model fits obtained with INST-MFA. This article is protected by copyright. All rights reserved.

This is a handcraft: valuation, morality, and the social meanings of payments for psychoanalysis

This article examines valuation and payment practices of psychoanalysts in Buenos Aires, Argentina. Psychoanalysts do not use explicit sliding scales but rather reach an agreement about fees in conversation with the patient. This negotiation is conducted with some principles of gift-giving, where parties try to give more, rather than through competitive bargaining (an inverted bazaar). Drawing on the sociology of money, morals and markets, and valuation studies literatures, I distinguish four factors to explain this: 1) Some formally produced prices as well as market mechanisms shape benchmarks for fees, but the peculiar service psychologists offer (which makes quality judgments hard), the way patients and therapists are matched, and the lack of public information about prices allow for high flexibility in price-setting; these are structural factors that remain unsaid in the conversation on fees. 2) A professional narrative that highlights a responsibility towards patients that should not be contaminated by economic interest. 3) Psychoanalysts' elaborations on the meanings of the payment, which should reflect the uniqueness of each patient and the bond analyst-patient and symbolize the patient's commitment to treatment, involving a cost and a loss beyond the economic. 4) The prevalence of cash, face-to-face payment without intermediaries, which helps desacralize the analyst and disentangle the session from the rest of the economic life of the analyst, but impedes evading moralization of the transaction. Payments in psychoanalysis are delicate arrangements, and analysts often stress about valuation and payments. They have to be careful to ensure this flexibility results in morally acceptable transactions.

Predicting population health with machine learning: a scoping review

OBJECTIVE

To determine how machine learning has been applied to prediction applications in population health contexts. Specifically, to describe which outcomes have been studied, the data sources most widely used and whether reporting of machine learning predictive models aligns with established reporting guidelines.

DESIGN

A scoping review.

DATA SOURCES

MEDLINE, EMBASE, CINAHL, ProQuest, Scopus, Web of Science, Cochrane Library, INSPEC and ACM Digital Library were searched on 18 July 2018.

ELIGIBILITY CRITERIA

We included English articles published between 1980 and 2018 that used machine learning to predict population-health-related outcomes. We excluded studies that only used logistic regression or were restricted to a clinical context.

DATA EXTRACTION AND SYNTHESIS

We summarised findings extracted from published reports, which included general study characteristics, aspects of model development, reporting of results and model discussion items.

RESULTS

Of 22 618 articles found by our search, 231 were included in the review. The USA (n=71, 30.74%) and China (n=40, 17.32%) produced the most studies. Cardiovascular disease (n=22, 9.52%) was the most studied outcome. The median number of observations was 5414 (IQR=16 543.5) and the median number of features was 17 (IQR=31). Health records (n=126, 54.5%) and investigator-generated data (n=86, 37.2%) were the most common data sources. Many studies did not incorporate recommended guidelines on machine learning and predictive modelling. Predictive discrimination was commonly assessed using area under the receiver operator curve (n=98, 42.42%) and calibration was rarely assessed (n=22, 9.52%).

CONCLUSIONS

Machine learning applications in population health have concentrated on regions and diseases well represented in traditional data sources, infrequently using big data. Important aspects of model development were under-reported. Greater use of big data and reporting guidelines for predictive modelling could improve machine learning applications in population health.

REGISTRATION NUMBER

Registered on the Open Science Framework on 17 July 2018 (available at https://osf.io/rnqe6/).

Long-term outcome of postmenopausal women with non-atypical endometrial hyperplasia on endometrial sampling

OBJECTIVE

To assess the long-term outcome of postmenopausal women diagnosed with non-atypical endometrial hyperplasia (NEH).

METHODS

This was a retrospective study of women aged 55 or older who underwent endometrial sampling in our academic medical center between 1997 and 2008. Women who had a current or recent (< 2 years) histological diagnosis of NEH were included in the study group and were compared with those diagnosed with atrophic endometrium (AE). Outcome data were obtained until February 2018. The main outcomes were risk of progression to endometrial carcinoma and risk of persistence, recurrence or new development of endometrial hyperplasia (EH) ('persistent EH'). Logistic regression analysis was used to identify covariates that were independent risk factors for progression to endometrial cancer or persistent EH.

RESULTS

During the study period, 1808 women aged 55 or older underwent endometrial sampling. The median surveillance time was 10.0 years. Seventy-two women were found to have a current or recent diagnosis of NEH and were compared with 722 women with AE. When compared to women with AE, women with NEH had significantly higher body mass index (33.9 kg/m² vs 30.6 kg/m² ; P = 0.01), greater endometrial thickness (10.00 mm vs 6.00 mm; P = 0.01) and higher rates of progression to type-1 endometrial cancer (8.3% vs 0.8%; P = 0.0003) and persistent NEH (22.2% vs 0.7%; P < 0.0001). They also had a higher rate of progression to any type of uterine cancer or persistent EH (33.3% vs 3.5%; P < 0.0001). Women with NEH had a significantly higher rate of future surgical intervention (51.4% vs 15.8%; P < 0.0001), including future hysterectomy (34.7% vs 9.8%; P < 0.0001). On multivariable logistic regression analysis, only NEH remained a significant risk factor for progression to endometrial cancer or persistence of EH.

CONCLUSIONS

Postmenopausal women with NEH are at significant risk for persistent EH and progression to endometrial cancer, at rates higher than those reported previously. Guidelines for the appropriate management of postmenopausal women with NEH are needed in order to decrease the rate of persistent disease or progression to cancer. Copyright © 2019 ISUOG. Published by John Wiley & Sons Ltd.

Uropathogen Resistance: Are Laboratory-Generated Data Reliable Enough?

Therapy of uncomplicated urinary tract infection (UTI) is based on knowledge of in vitro susceptibility profiles of uropathogens in the geographic region. Microbiological surveillance systems, which lack epidemiological and clinical data to differentiate between complicated and uncomplicated UTI may incorrectly estimate rates of resistance in the community. We determined the susceptibility profile of bacteria isolated from a random sample of 124 adult outpatients with diagnosis of uncomplicated UTI and we compared it with all outpatient urine specimens collected by the same participant laboratories during the same period. Escherichia coli was the most frequently isolated uropathogen in patients with uncomplicated UTI, and its rate of resistance to different antimicrobials was lower than overall resistance rates to E. coli reported by the participating laboratories during the same period. Resistance to cotrimoxazole was significantly lower. These results suggest that surveillance systems without clinical and epidemiological data may incorrectly gauge uropathogen resistance in the community.

In vitro assessment of antimicrobial peptides as potential agents against several oral bacteria

BACKGROUND

Antimicrobial peptides are components of the innate immunity that play an important role in systemic and oral health.

OBJECTIVES

The antibacterial activity of the amphibian-derived K4 -S4(1-15)a antimicrobial peptide was tested against oral pathogens associated with caries and periodontitis and compared with the activities of the human-derived antimicrobial peptides LL-37 and dhvar4a.

METHODS

Growth inhibition of planktonic bacteria was tested using standard microdilution assays. Live/Dead staining followed by confocal scanning laser microscopy (CSLM) was used to determine the bactericidal effect of K4 -S4(1-15)a on Streptococcus mutans attached to a glass surface or grown as biofilm.

RESULTS

The cariogenic species S. mutans, Streptococcus sobrinus, Lactobacillus paracasei and Actinomyces viscosus were resistant to LL-37 found in the oral cavity. Porphyromonas gingivalis was the species most resistant to the three tested peptides. K4 -S4(1-15)a demonstrated the highest activity against the tested planktonic bacteria. In addition, K4 -S4(1-15)a was bactericidal to surface-attached S. mutans as well as to S. mutans biofilms grown in vitro. However, surface attachment increased S. mutans resistance to the antimicrobial peptide.

CONCLUSIONS

Our results support growing evidence suggesting the use of antimicrobial peptides for prevention and treatment of oral disease.

Cetuximab shows activity in colorectal cancer patients with tumors that do not express the epidermal growth factor receptor by immunohistochemistry

PURPOSE

To establish evidence of activity, or lack thereof, of cetuximab-based therapy in patients with refractory colorectal cancer with tumors that do not demonstrate epidermal growth factor receptor (EGFR) expression by immunohistochemistry (IHC).

PATIENTS AND METHODS

Pharmacy computer records were reviewed to identify all patients who received cetuximab at Memorial Sloan-Kettering Cancer Center in a nonstudy setting during the first 3 months of cetuximab's commercial availability. Medical records of these patients were then reviewed to identify colorectal cancer patients who had experienced failure with a prior irinotecan-based regimen and who had a pathology report indicating an EGFR-negative tumor by IHC. Pathology slides from these patients were reviewed by a reference pathologist to confirm EGFR negativity, and computed tomography scans during cetuximab-based therapy were reviewed by a reference radiologist. Response rates were reported using WHO criteria.

RESULTS

Sixteen chemotherapy-refractory, EGFR-negative colorectal cancer patients who received cetuximab in a nonstudy setting were identified. Fourteen of these patients received cetuximab plus irinotecan, and two received cetuximab monotherapy. In the 16 patients, four major objective responses were seen (response rate, 25%; 95% CI, 4% to 46%).

CONCLUSION

Colorectal cancer patients with EGFR-negative tumors have the potential to respond to cetuximab-based therapies. EGFR analysis by current IHC techniques does not seem to have predictive value, and selection or exclusion of patients for cetuximab therapy on the basis of currently available EGFR IHC does not seem warranted.