Predicting first time depression onset in pregnancy: applying machine learning methods to patient-reported data.

PURPOSE: To develop a machine learning algorithm, using patient-reported data from early pregnancy, to predict later onset of first time moderate-to-severe depression. METHODS: A sample of 944 U.S. patient participants from a larger longitudinal observational cohortused a prenatal support mobile app from September 2019 to April 2022. Participants self-reported clinical and social risk factors during first trimester initiation of app use and completed voluntary depression screenings in each trimester. Several machine learning algorithms were applied to self-reported data, including a novel algorithm for causal discovery. Training and test datasets were built from a randomized 80/20 data split. Models were evaluated on their predictive accuracy and their simplicity (i.e., fewest variables required for prediction). RESULTS: Among participants, 78% identified as white with an average age of 30 [IQR 26-34]; 61% had income ≥ $50,000; 70% had a college degree or higher; and 49% were nulliparous. All models accurately predicted first time moderate-severe depression using first trimester baseline data (AUC 0.74-0.89, sensitivity 0.35-0.81, specificity 0.78-0.95). Several predictors were common across models, including anxiety history, partnered status, psychosocial factors, and pregnancy-specific stressors. The optimal model used only 14 (26%) of the possible variables and had excellent accuracy (AUC = 0.89, sensitivity = 0.81, specificity = 0.83). When food insecurity reports were included among a subset of participants, demographics, including race and income, dropped out and the model became more accurate (AUC = 0.93) and simpler (9 variables). CONCLUSION: A relatively small amount of self-report data produced a highly predictive model of first time depression among pregnant individuals.

Development of risk prediction models for preterm delivery in a rural setting in Ethiopia.

Background: Preterm birth complications are the leading causes of death among children under five years. However, the inability to accurately identify pregnancies at high risk of preterm delivery is a key practical challenge, especially in resource-constrained settings with limited availability of biomarkers assessment. Methods: We evaluated whether risk of preterm delivery can be predicted using available data from a pregnancy and birth cohort in Amhara region, Ethiopia. All participants were enrolled in the cohort between December 2018 and March 2020. The study outcome was preterm delivery, defined as any delivery occurring before week 37 of gestation regardless of vital status of the foetus or neonate. A range of sociodemographic, clinical, environmental, and pregnancy-related factors were considered as potential inputs. We used Cox and accelerated failure time models, alongside decision tree ensembles to predict risk of preterm delivery. We estimated model discrimination using the area-under-the-curve (AUC) and simulated the conditional distributions of cervical length (CL) and foetal fibronectin (FFN) to ascertain whether they could improve model performance. Results: We included 2493 pregnancies; among them, 138 women were censored due to loss-to-follow-up before delivery. Overall, predictive performance of models was poor. The AUC was highest for the tree ensemble classifier (0.60, 95% confidence interval = 0.57-0.63). When models were calibrated so that 90% of women who experienced a preterm delivery were classified as high risk, at least 75% of those classified as high risk did not experience the outcome. The simulation of CL and FFN distributions did not significantly improve models' performance. Conclusions: Prediction of preterm delivery remains a major challenge. In resource-limited settings, predicting high-risk deliveries would not only save lives, but also inform resource allocation. It may not be possible to accurately predict risk of preterm delivery without investing in novel technologies to identify genetic factors, immunological biomarkers, or the expression of specific proteins.

Development of Prediction Models for Antenatal Care Attendance in Amhara Region, Ethiopia.

Importance: Antenatal care prevents maternal and neonatal deaths and improves birth outcomes. There is a lack of predictive models to identify pregnant women who are at high risk of failing to attend antenatal care in low-resource settings. Objective: To develop a series of predictive models to identify women who are at high risk of failing to attend antenatal care in a rural setting in Ethiopia. Design, Setting, and Participants: This prognostic study used data from the Birhan Health and Demographic Surveillance System and its associated pregnancy and child cohort. The study was conducted at the Birhan field site, North Shewa zone, Ethiopia, a platform for community- and facility-based research and training, with a focus on maternal and child health. Participants included women enrolled during pregnancy in the pregnancy and child cohort between December 2018 and March 2020, who were followed-up in home and facility visits. Data were analyzed from April to December 2022. Exposures: A wide range of sociodemographic, economic, medical, environmental, and pregnancy-related factors were considered as potential predictors. The selection of potential predictors was guided by literature review and expert knowledge. Main Outcomes and Measures: The outcome of interest was failing to attend at least 1 antenatal care visit during pregnancy. Prediction models were developed using logistic regression with regularization via the least absolute shrinkage and selection operator and ensemble decision trees and assessed using the area under the receiving operator characteristic curve (AUC). Results: The study sample included 2195 participants (mean [SD] age, 26.8 [6.1] years; mean [SD] gestational age at enrolment, 25.5 [8.8] weeks). A total of 582 women (26.5%) failed to attend antenatal care during cohort follow-up. The AUC was 0.61 (95% CI, 0.58-0.64) for the regularized logistic regression model at conception, with higher values for models predicting at weeks 13 (AUC, 0.68; 95% CI, 0.66-0.71) and 24 (AUC, 0.66; 95% CI, 0.64-0.69). AUC values were similar with slightly higher performance for the ensembles of decision trees (conception: AUC, 0.62; 95% CI, 0.59-0.65; 13 weeks: AUC, 0.70; 95% CI, 0.67-0.72; 24 weeks: AUC, 0.67; 95% CI, 0.64-0.69). Conclusions and Relevance: This prognostic study presents a series of prediction models for antenatal care attendance with modest performance. The developed models may be useful to identify women at high risk of missing their antenatal care visits to target interventions to improve attendance rates. This study opens the possibility to develop and validate easy-to-use tools to project health-related behaviors in settings with scarce resources.

A Peer-Led, Artificial Intelligence-Augmented Social Network Intervention to Prevent HIV Among Youth Experiencing Homelessness.

BACKGROUND: Youth experiencing homelessness (YEH) are at elevated risk of HIV/AIDS and disproportionately identify as racial, ethnic, sexual, and gender minorities. We developed a new peer change agent (PCA) HIV prevention intervention with 3 arms: (1) an arm using an artificial intelligence (AI) planning algorithm to select PCAs; (2) a popularity arm, the standard PCA approach, operationalized as highest degree centrality (DC); and (3) an observation-only comparison group. SETTING: A total of 713 YEH were recruited from 3 drop-in centers in Los Angeles, CA. METHODS: Youth consented and completed a baseline survey that collected self-reported data on HIV knowledge, condom use, and social network information. A quasi-experimental pretest/posttest design was used; 472 youth (66.5% retention at 1 month postbaseline) and 415 youth (58.5% retention at 3 months postbaseline) completed follow-up. In each intervention arm (AI and DC), 20% of youth was selected as PCAs and attended a 4-hour initial training, followed by 7 weeks of half-hour follow-up sessions. Youth disseminated messages promoting HIV knowledge and condom use. RESULTS: Using generalized estimating equation models, there was a significant reduction over time (P < 0.001) and a significant time by AI arm interaction (P < 0.001) for condomless anal sex act. There was a significant increase in HIV knowledge over time among PCAs in DC and AI arms. CONCLUSIONS: PCA models that promote HIV knowledge and condom use are efficacious for YEH. Youth are able to serve as a bridge between interventionists and their community. Interventionists should consider working with computer scientists to solve implementation problems.

Test sensitivity is secondary to frequency and turnaround time for COVID-19 screening.

The COVID-19 pandemic has created a public health crisis. Because SARS-CoV-2 can spread from individuals with presymptomatic, symptomatic, and asymptomatic infections, the reopening of societies and the control of virus spread will be facilitated by robust population screening, for which virus testing will often be central. After infection, individuals undergo a period of incubation during which viral titers are too low to detect, followed by exponential viral growth, leading to peak viral load and infectiousness and ending with declining titers and clearance. Given the pattern of viral load kinetics, we model the effectiveness of repeated population screening considering test sensitivities, frequency, and sample-to-answer reporting time. These results demonstrate that effective screening depends largely on frequency of testing and speed of reporting and is only marginally improved by high test sensitivity. We therefore conclude that screening should prioritize accessibility, frequency, and sample-to-answer time; analytical limits of detection should be secondary.

Modeling between-population variation in COVID-19 dynamics in Hubei, Lombardy, and New York City.

As the COVID-19 pandemic continues, formulating targeted policy interventions that are informed by differential severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission dynamics will be of vital importance to national and regional governments. We develop an individual-level model for SARS-CoV-2 transmission that accounts for location-dependent distributions of age, household structure, and comorbidities. We use these distributions together with age-stratified contact matrices to instantiate specific models for Hubei, China; Lombardy, Italy; and New York City, United States. Using data on reported deaths to obtain a posterior distribution over unknown parameters, we infer differences in the progression of the epidemic in the three locations. We also examine the role of transmission due to particular age groups on total infections and deaths. The effect of limiting contacts by a particular age group varies by location, indicating that strategies to reduce transmission should be tailored based on population-specific demography and social structure. These findings highlight the role of between-population variation in formulating policy interventions. Across the three populations, though, we find that targeted "salutary sheltering" by 50% of a single age group may substantially curtail transmission when combined with the adoption of physical distancing measures by the rest of the population.

Test sensitivity is secondary to frequency and turnaround time for COVID-19 surveillance.

The COVID-19 pandemic has created a public health crisis. Because SARS-CoV-2 can spread from individuals with pre-symptomatic, symptomatic, and asymptomatic infections, the re-opening of societies and the control of virus spread will be facilitated by robust surveillance, for which virus testing will often be central. After infection, individuals undergo a period of incubation during which viral titers are usually too low to detect, followed by an exponential viral growth, leading to a peak viral load and infectiousness, and ending with declining viral levels and clearance. Given the pattern of viral load kinetics, we model surveillance effectiveness considering test sensitivities, frequency, and sample-to-answer reporting time. These results demonstrate that effective surveillance depends largely on frequency of testing and the speed of reporting, and is only marginally improved by high test sensitivity. We therefore conclude that surveillance should prioritize accessibility, frequency, and sample-to-answer time; analytical limits of detection should be secondary.

Inferring individual-level processes from population-level patterns in cultural evolution.

Our species is characterized by a great degree of cultural variation, both within and between populations. Understanding how group-level patterns of culture emerge from individual-level behaviour is a long-standing question in the biological and social sciences. We develop a simulation model capturing demographic and cultural dynamics relevant to human cultural evolution, focusing on the interface between population-level patterns and individual-level processes. The model tracks the distribution of variants of cultural traits across individuals in a population over time, conditioned on different pathways for the transmission of information between individuals. From these data, we obtain theoretical expectations for a range of statistics commonly used to capture population-level characteristics (e.g. the degree of cultural diversity). Consistent with previous theoretical work, our results show that the patterns observed at the level of groups are rooted in the interplay between the transmission pathways and the age structure of the population. We also explore whether, and under what conditions, the different pathways can be distinguished based on their group-level signatures, in an effort to establish theoretical limits to inference. Our results show that the temporal dynamic of cultural change over time retains a stronger signature than the cultural composition of the population at a specific point in time. Overall, the results suggest a shift in focus from identifying the one individual-level process that likely produced the observed data to excluding those that likely did not. We conclude by discussing the implications for empirical studies of human cultural evolution.

Altruists Proliferate Even at a Selective Disadvantage within Their Own Niche.

The evolutionary origin of altruism is a long-standing puzzle. Numerous explanations have been proposed, most prominently based on inclusive fitness or group selection. One possibility that has not yet been considered is that new niches will be created disproportionately often when altruism appears, perhaps by chance, causing altruists to be over-represented in such new niches. This effect is a novel variant of group selection in which altruistic groups benefit by discovering unoccupied niches instead of by competing for the limited resources within a single niche. Both an analytical population genetics model and computational simulations support that altruism systematically arises due to this side effect of increased carrying capacity even when it is strongly selected against within any given niche. In fact, even when selection is very strongly negative and altruism does not develop in most populations, it can still be expected to be observed in a consistent fraction of species. The ecological structure provided by niches thereby may be sufficient for altruists to proliferate even if they are always at a disadvantage within each niche considered individually.

Inference of Cultural Transmission Modes Based on Incomplete Information.

In this article we explore the theoretical limits of the inference of cultural transmission modes based on sparse population-level data. We approach this problem by investigating whether different transmission modes produce different temporal dynamics of cultural change. In particular, we explore whether different transmission modes result in sufficiently different distributions of the average time a variant stays the most common variant in the population, tmax, so that their inference can be guaranteed on the basis of an estimate of tmax. We assume time series data detailing the frequencies of different variants of a cultural trait in a population at different points in time and investigate the temporal resolution (i.e., the length of the time series and the distance between consecutive time points) that is needed to ensure distinguishability between transmission modes. We find that under complete information most transmission modes can be distinguished on the basis of the statistic tmax; however, we should not expect the same results if only infrequent information about the most common cultural variant in the population is available.