The impact of methodological choices when developing predictive models using urinary metabolite data.

The continuous evolution of metabolomics over the past two decades has stimulated the search for metabolic biomarkers of many diseases. Metabolomic data measured from urinary samples can provide rich information of the biological events triggered by organ rejection in pediatric kidney transplant recipients. With additional validation, metabolic markers can be used to build clinically useful diagnostic tools. However, there are many methodological steps ranging from data processing to modeling that can influence the performance of the resulting metabolomic classifiers. In this study we focus on the comparison of various classification methods that can handle the complex structure of metabolomic data, including regularized classifiers, partial least squares discriminant analysis, and nonlinear classification models. We also examine the effectiveness of a physiological normalization technique widely used in the clinical and biochemical literature but not extensively analyzed and compared in urine metabolomic studies. While the main objective of this work is to interrogate metabolomic data of pediatric kidney transplant recipients to improve the diagnosis of T cell-mediated rejection (TCMR), we also analyze three independent datasets from other disease conditions to investigate the generalizability of our findings.

Spatiotemporal patterns of paralytic shellfish toxins and their relationships with environmental variables in British Columbia, Canada from 2002 to 2012.

BACKGROUND: Harmful algal blooms produce paralytic shellfish toxins that accumulate in the tissues of filter feeding shellfish. Ingestion of these toxic shellfish can cause a serious and potentially fatal condition known as paralytic shellfish poisoning (PSP). The coast of British Columbia is routinely monitored for shellfish toxicity, and this study uses data from the monitoring program to identify spatiotemporal patterns in shellfish toxicity events and their relationships with environmental variables. METHODS: The dinoflagellate genus Alexandrium produces the most potent paralytic shellfish toxin, saxitoxin (STX). Data on all STX measurements were obtained from 49 different shellfish monitoring sites along the coast of British Columbia for 2002-2012, and monthly toxicity events were identified. We performed hierarchical cluster analysis to group sites that had events in similar areas with similar timing. Machine learning techniques were used to model the complex relationships between toxicity events and environmental variables in each group. RESULTS: The Strait of Georgia and the west coast of Vancouver Island had unique toxicity regimes. Out of the seven environmental variables used, toxicity in each cluster could be described by multivariable models including monthly sea surface temperature, air temperature, sea surface salinity, freshwater discharge, upwelling, and photosynthetically active radiation. The sea surface salinity and freshwater discharge variables produced the strongest univariate models for both geographic areas. CONCLUSIONS: Applying these methods in coastal regions could allow for the prediction of shellfish toxicity events by environmental conditions. This has the potential to optimize biotoxin monitoring, improve public health surveillance, and engage the shellfish industry in helping to reduce the risk of PSP.

A proposed case-control framework to probabilistically classify individual deaths as expected or excess during extreme hot weather events.

BACKGROUND: Most excess deaths that occur during extreme hot weather events do not have natural heat recorded as an underlying or contributing cause. This study aims to identify the specific individuals who died because of hot weather using only secondary data. A novel approach was developed in which the expected number of deaths was repeatedly sampled from all deaths that occurred during a hot weather event, and compared with deaths during a control period. The deaths were compared with respect to five factors known to be associated with hot weather mortality. Individuals were ranked by their presence in significant models over 100 trials of 10,000 repetitions. Those with the highest rankings were identified as probable excess deaths. Sensitivity analyses were performed on a range of model combinations. These methods were applied to a 2009 hot weather event in greater Vancouver, Canada. RESULTS: The excess deaths identified were sensitive to differences in model combinations, particularly between univariate and multivariate approaches. One multivariate and one univariate combination were chosen as the best models for further analyses. The individuals identified by multiple combinations suggest that marginalized populations in greater Vancouver are at higher risk of death during hot weather. CONCLUSIONS: This study proposes novel methods for classifying specific deaths as expected or excess during a hot weather event. Further work is needed to evaluate performance of the methods in simulation studies and against clinically identified cases. If confirmed, these methods could be applied to a wide range of populations and events of interest.

Municipal contraceptive services, socioeconomic status and teenage pregnancy in Finland: a longitudinal study.

OBJECTIVES: Declining teenage pregnancy rates have been linked to improved access to youth-friendly contraceptive services, but information on the combined association of these services and socioeconomic factors with teenage pregnancy is lacking. DESIGN AND SETTING: This retrospective longitudinal register-based study covers the annual teenage childbirth and induced abortion rates in the 100 largest municipalities in Finland in 2000-2018. We investigated the combined association of regional, socioeconomic (ie, education level and need for social assistance) and adolescent contraceptive service variables (ie, free-of-charge contraception, an adolescent-only clinic and availability of over-the-counter emergency contraception (OTC EC)) with teenage childbirth and induced abortion rates at the municipality level by using Poisson mixed-effects model. PRIMARY OUTCOME MEASURES: Annual teenage childbirth and induced abortion rates as numbers per 1000 teenage girls aged 15-19 years old in the 100 largest municipalities in Finland from 2000 to 2018. RESULTS: The following variables were significantly associated with both lower teenage childbirth and induced abortion rates when adjusted for all the other variables used in the model: providing free-of-charge contraception (rate ratio (RR) 0.82 (95% CI 0.73 to 0.92) and RR 0.87 (95% CI 0.79 to 0.96), respectively), availability of OTC EC without age limit (RR 0.70 (95% CI 0.67 to 0.75) and RR 0.74 (95% CI 0.71 to 0.78), respectively), and high education level of the municipality (RR 0.94 (95% CI 0.94 to 0.95) and RR 0.94 (95% CI 0.93 to 0.94), respectively). CONCLUSION: Providing free-of-charge contraception and availability of OTC EC without age limit are associated with lower teenage pregnancy rates. These services combined with proper counselling are thus important contents of youth-friendly contraceptive services that should be provided equally for all teenagers in order to further reduce teenage pregnancy rates.

The Heat Exposure Integrated Deprivation Index (HEIDI): A data-driven approach to quantifying neighborhood risk during extreme hot weather.

Mortality attributable to extreme hot weather is a growing concern in many urban environments, and spatial heat vulnerability indexes are often used to identify areas at relatively higher and lower risk. Three indexes were developed for greater Vancouver, Canada using a pool of 20 potentially predictive variables categorized to reflect social vulnerability, population density, temperature exposure, and urban form. One variable was chosen from each category: an existing deprivation index, senior population density, apparent temperature, and road density, respectively. The three indexes were constructed from these variables using (1) unweighted, (2) weighted, and (3) data-driven Heat Exposure Integrated Deprivation Index (HEIDI) approaches. The performance of each index was assessed using mortality data from 1998-2014, and the maps were compared with respect to spatial patterns identified. The population-weighted spatial correlation between the three indexes ranged from 0.68-0.89. The HEIDI approach produced a graduated map of vulnerability, whereas the other approaches primarily identified areas of highest risk. All indexes performed best under extreme temperatures, but HEIDI was more useful at lower thresholds. Each of the indexes in isolation provides valuable information for public health protection, but combining the HEIDI approach with unweighted and weighted methods provides richer information about areas most vulnerable to heat.