Identifying infants at risk of sudden unexpected death with an automated predictive risk model.

BACKGROUND/OBJECTIVE: Sudden unexpected infant death (SUID) is a common cause of infant death. We evaluated whether a predictive risk model (PRM) - Hello Baby - which was developed to stratify children by risk of entry into foster care could also identify infants at highest risk of SUID and non-fatal unsafe sleep events. PARTICIPANTS AND SETTING: Cases: Infants with SUID or an unsafe sleep event over 5½ years in a single county. CONTROLS: All births in the same county. METHODS: Retrospective case-control study. Demographic and clinical data were collected and a Hello Baby PRM score was assigned. Descriptive statistics and the predictive value of a PRM score of 20 were calculated. RESULTS: Infants with SUID (n = 62) or an unsafe sleep event (n = 37) (cases) were compared with 23,366 births (controls). Cases and controls were similar for all demographic and clinical data except that infants with unsafe sleep events were older. Median PRM score for cases was higher than controls (17.5 vs. 10, p < 0.001); 50 % of cases had a PRM score 17-20 vs. 16 % of controls (p < 0.001). CONCLUSIONS: The Hello Baby PRM can identify newborns at high risk of SUID and non-fatal unsafe sleep events. The ability to identify high-risk newborns prior to a negative outcome allows for individualized evaluation of high-risk families for modifiable risk factors which are potentially amenable to intervention. This approach is limited by the fact that not all counties can calculate a PRM or similar score automatically.

Using predictive risk modelling to identify patients with hidden health needs in an Aboriginal and Torres Strait Islander health service.

BACKGROUND AND OBJECTIVES: In partnership with an Aboriginal and Torres Strait Islander community-controlled health service, we explored the use of a machine learning tool to identify high-needs patients for whom services are harder to reach and, hence, who do not engage with primary care. METHOD: Using deidentified electronic health record data, two predictive risk models (PRMs) were developed to identify patients who were: (1) unlikely to have health checks as an indicator of not engaging with care; and (2) likely to rate their wellbeing as poor, as a measure of high needs. RESULTS: According to the standard metrics, the PRMs were good at predicting health checks but showed low reliability for detecting poor wellbeing. DISCUSSION: Results and feedback from clinicians were encouraging. With additional refinement, informed by clinic staff feedback, a deployable model should be feasible.

Using electronic health record data to predict future self-harm or suicidal ideation in young people treated by child and youth mental health services.

INTRODUCTION: Identifying young people who are at risk of self-harm or suicidal ideation (SHoSI) is a priority for mental health clinicians. We explore the utility of routinely collected data in developing a tool to aid early identification of those at risk. METHOD: We used electronic health records of 4610 young people aged 5-19 years who were treated by Child and Youth Mental Health Services (CYMHS) in greater Brisbane, Australia. Two Lasso models were trained to predict the risk of future SHoSI in young people currently rated SHoSI; and those who were not. RESULTS: For currently non-SHoSI children, an Area Under the Receiver Operating Characteristics (AUC) of 0.78 was achieved. Those with the highest risk were 4.97 (CI 4.35-5.66) times more likely to be categorized as SHoSI in the future. For current SHoSI children, the AUC was 0.62. CONCLUSION: A prediction model with fair overall predictive power for currently non-SHoSI children was generated. Predicting persistence for SHoSI was more difficult. The electronic health records alone were not sufficient to discriminate at acceptable levels and may require adding unstructured data such as clinical notes. To optimally predict SHoSI models need to be tested and validated separately for those young people with varying degrees of risk.

Machine learning to predict poor school performance in paediatric survivors of intensive care: a population-based cohort study.

PURPOSE: Whilst survival in paediatric critical care has improved, clinicians lack tools capable of predicting long-term outcomes. We developed a machine learning model to predict poor school outcomes in children surviving intensive care unit (ICU). METHODS: Population-based study of children < 16 years requiring ICU admission in Queensland, Australia, between 1997 and 2019. Failure to meet the National Minimum Standard (NMS) in the National Assessment Program-Literacy and Numeracy (NAPLAN) assessment during primary and secondary school was the primary outcome. Routine ICU information was used to train machine learning classifiers. Models were trained, validated and tested using stratified nested cross-validation. RESULTS: 13,957 childhood ICU survivors with 37,200 corresponding NAPLAN tests after a median follow-up duration of 6 years were included. 14.7%, 17%, 15.6% and 16.6% failed to meet NMS in school grades 3, 5, 7 and 9. The model demonstrated an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.8 (standard deviation SD, 0.01), with 51% specificity to reach 85% sensitivity [relative Area Under the Precision Recall Curve (rel-AUPRC) 3.42, SD 0.06]. Socio-economic status, illness severity, and neurological, congenital, and genetic disorders contributed most to the predictions. In children with no comorbidities admitted between 2009 and 2019, the model achieved a AUROC of 0.77 (SD 0.03) and a rel-AUPRC of 3.31 (SD 0.42). CONCLUSIONS: A machine learning model using data available at time of ICU discharge predicted failure to meet minimum educational requirements at school age. Implementation of this prediction tool could assist in prioritizing patients for follow-up and targeting of rehabilitative measures.

Estimating the economic costs of Indigenous health inequities in New Zealand: a retrospective cohort analysis.

OBJECTIVES: Despite significant international interest in the economic impacts of health inequities, few studies have quantified the costs associated with unfair and preventable ethnic/racial health inequities. This Indigenous-led study is the first to investigate health inequities between Maori and non-Maori adults in New Zealand (NZ) and estimate the economic costs associated with these differences. DESIGN: Retrospective cohort analysis. Quantitative epidemiological methods and 'cost-of-illness' (COI) methodology were employed, within a Kaupapa Maori theoretical framework. SETTING: Data for 2003-2014 were obtained from national data collections held by NZ government agencies, including hospitalisations, mortality, outpatient and primary care consultations, laboratory and pharmaceutical usage and accident claims. PARTICIPANTS: All adults in NZ aged 15 years and above who had engagement with the health system between 2003 and 2014 (deidentified). PRIMARY AND SECONDARY OUTCOME MEASURES: Rates of 'potentially avoidable' hospitalisations and mortality as well as 'excess or underutilisation' of healthcare were calculated, as the difference between actual rates for Maori and the rate expected if Maori had the same rates as non-Maori. These differences were then quantified using COI methodology to estimate the financial cost of ethnic inequities. RESULTS: In this conservative estimate, health inequities between Maori and non-Maori adults cost NZ$863.3 million per year. Direct costs of NZ$39.9 million per year included costs from ambulatory sensitive hospitalisations and outpatient care, with cost savings from underutilisation of primary care. Indirect costs of NZ$823.4 million per year came from years of life lost and lost wages. CONCLUSIONS: Indigenous adult health inequities in NZ create significant direct and indirect costs. The 'cost of doing nothing' is predominantly borne by Indigenous communities and society. The net cost of adult health inequities to the government conceals substantial savings to the government from underutilisation of primary care and accident/injury care.

Sibyl: Understanding and Addressing the Usability Challenges of Machine Learning In High-Stakes Decision Making.

Machine learning (ML) is being applied to a diverse and ever-growing set of domains. In many cases, domain experts - who often have no expertise in ML or data science - are asked to use ML predictions to make high-stakes decisions. Multiple ML usability challenges can appear as result, such as lack of user trust in the model, inability to reconcile human-ML disagreement, and ethical concerns about oversimplification of complex problems to a single algorithm output. In this paper, we investigate the ML usability challenges that present in the domain of child welfare screening through a series of collaborations with child welfare screeners. Following the iterative design process between the ML scientists, visualization researchers, and domain experts (child screeners), we first identified four key ML challenges and honed in on one promising explainable ML technique to address them (local factor contributions). Then we implemented and evaluated our visual analytics tool, Sibyl, to increase the interpretability and interactivity of local factor contributions. The effectiveness of our tool is demonstrated by two formal user studies with 12 non-expert participants and 13 expert participants respectively. Valuable feedback was collected, from which we composed a list of design implications as a useful guideline for researchers who aim to develop an interpretable and interactive visualization tool for ML prediction models deployed for child welfare screeners and other similar domain experts.

Measuring the working experience of doctors in training.

Using an online tool, we report the association between tasks and 'affect' (underlying experience of feeling, emotion or mood) among 565 doctors in training, how positive and negative emotional intensity are associated with time of day, the extent to which positive affect is associated with breaks, and consideration about leaving the profession. Respondents spent approximately 25% of their day on paperwork or clinical work that did not involve patients, resulting in more negative emotions. Positive emotions were expressed for breaks, staff meetings, research, learning and clinical tasks that involved patients. Those having considered leaving the profession report more negative feelings. Systematic workplace changes (regular breaks, reducing paperwork and improved IT systems) could contribute to positive workday experiences and reduce intention to quit. Educators and employers have important roles in recognising, advocating for and implementing improvements at work to enhance wellbeing with potential to improve retention of doctors in training.

Hospital Injury Encounters of Children Identified by a Predictive Risk Model for Screening Child Maltreatment Referrals: Evidence From the Allegheny Family Screening Tool.

Importance: Nearly 6 million children are reported as allegedly experiencing abuse or neglect in the US annually. Child protection agencies are increasingly turning to automated predictive risk models (PRMs) that mine information found in routinely collected administrative data and estimate a likelihood that an individual will experience some future adverse outcome. Objective: To test if a PRM used at the time of referral for alleged maltreatment, which automatically generates a risk stratification score indicating the relative likelihood of future foster care placement, is also predictive of injury hospitalization data. Design, Setting, and Participants: This retrospective cohort study based on a probabilistic association between child protection and hospital encounter data was conducted in Allegheny County, Pennsylvania, and at Children's Hospital of Pittsburgh (Pittsburgh, Pennsylvania). Participants included children referred for alleged neglect or abuse in Allegheny County between April 1, 2010, and May 4, 2016. Exposures: Risk score generated from the PRM. Main Outcomes and Measures: Medical encounters (emergency department and inpatient hospitalizations) for any-cause injuries, suicide or self-inflicted harm injuries, and abuse injuries between 2002 and 2015 for children classified by the PRM to different risk levels at the time of a maltreatment referral. Cancer encounters were used as a placebo test. Results: Of 47 305 participants, 23 601 (49.9%) were girls, the mean (SD) age at referral was 8 (5.7) years, 28 211 (59.6%) were black, and 19 094 (40.4%) were nonblack. Children who scored in the highest 5% risk group by the PRM were more likely to have a medical encounter for an injury during the follow-up period than low-risk children (ie, those in the bottom 50% of risk). Specifically, among children referred for maltreatment and classified as highest risk, the rate of experiencing an any-cause injury encounter was 14.5 (95% CI, 13.1-15.9) per 100 compared with children who scored as low risk who had an any-cause injury encounter rate of 4.9 (95% CI, 4.7-5.2) per 100. For abuse-associated injury encounters, the rate for high-risk children was 2.0 (95% CI, 1.5-2.6) per 100 and that of low-risk children was 0.2 (95% CI, 0.2-0.3) per 100; for suicide and self-harm, the high-risk encounter rate was 1.0 (95% CI, 0.6-1.4) per 100 and that of low-risk children was 0.1 (95% CI, 0.1-0.1) per 100. There was no association between risk scores and cancer encounters. Conclusions and Relevance: Findings confirm that children reported for having experienced alleged maltreatment and classified by a PRM tool to be at high risk of foster care placement are also at increased risk of emergency department and in-patient hospitalizations for injuries.

Ethnic Disparities in Childhood Prevalence of Maltreatment: Evidence From a New Zealand Birth Cohort.

Objectives. To document ethnic disparities in childhood abuse and neglect among New Zealand children.Methods. We followed the 1998 New Zealand birth cohort of 56 904 children through 2016. We determined the cumulative childhood prevalence of reports to child protective services (CPS), substantiated maltreatment (by subtype), and out-of-home placements, from birth to age 18 years, by ethnic group. We also developed estimates stratified by maternal age and community deprivation levels.Results. We identified substantial ethnic differences in child maltreatment and child protection involvement. Both Maori and Pacific Islander children had a far greater likelihood of being reported to CPS, being substantiated as victims, and experiencing an out-of-home placement than other children. Across all levels of CPS interactions, rates of Maori involvement were more than twice those of Pacific Islander children and more than 3 times those of European children.Conclusions. Despite long-standing child support policies and reparation for breaches of Indigenous people's rights, significant child maltreatment disparities persist. More work is needed to understand how New Zealand's public benefit services can be more responsive to the needs of Indigenous families and their children.

Adverse childhood experiences and school readiness outcomes: results from the Growing Up in New Zealand study.

AIM: The Center for Disease Control's (CDC) Adverse Childhood Experiences (ACEs) have been associated with adverse health consequences in adults and children, but less is known about any association between ACE and early learning skills. We investigated the relationship between ACEs and objective preschool measures of skills using the Growing up In New Zealand (GUiNZ) cohort study (n=5,562; 2009-2015). METHODS: We mapped standard ACE definitions to GUiNZ to determine the prevalence of ACEs. We performed regression analysis to investigate the association between ACEs and a range of outcome measures, including counting up to 10, counting down from 10, letter recognition, affective knowledge, name writing, number writing and delayed gratification. RESULTS: Before entering primary school, 52.8% of GUiNZ children experienced at least one ACE. We found a dose-response relationship with seven of the eight tests. For example, after statistically adjusting for multiple potential confounders, for each one additional ACE, children were 1.12 times more likely to be unable to count up from 1-10 (95% Confidence Interval 1.04-1.19). CONCLUSIONS: Awareness of the negative impact of ACEs on school readiness should aid in the development and prioritisation of prevention strategies to reduce the occurrence and impact of ACEs in children.

Monthly spending dynamics of the elderly following a health shock: Evidence from Singapore.

We use novel longitudinal data from 19 monthly waves of the Singapore Life Panel to examine the short-term dynamics of the effects health shocks have on household health and nonhealth spending and income by the elderly. The health shocks we study are the occurrence of new major conditions such as cancer, heart problems, and minor conditions (e.g., diabetes and hypertension). Our empirical strategy is based on an event study approach that exploits unanticipated changes in health status through the diagnosis of new health conditions. We find that major shocks have large and persistent effects whereas minor shocks have small and mainly contemporaneous effects. We find that household income reduces following a major shock for males but not females. Major health shocks lead to a decrease in households' nonhealth expenditures that is particularly pronounced for cancer and stroke sufferers, driven largely by reductions in leisure spending. The financial impact of major shocks on medical saving account balances occurs to those without private health insurance, whereas the impact is on cash balances for privately insured individuals.

Estimating the economic costs of ethnic health inequities: protocol for a prevalence-based cost-of-illness study in New Zealand (2003-2014).

INTRODUCTION: There is significant international interest in the economic impacts of persistent inequities in morbidity and mortality. However, very few studies have quantified the costs associated with unfair and preventable ethnic/racial inequities in health. The proposed study will investigate inequities in health between the indigenous Maori and non-Maori adult population in New Zealand (15 years and older) and estimate the economic costs associated with these differences. METHODS AND ANALYSIS: The study will use national collections data that is held by government agencies in New Zealand including hospitalisations, mortality, outpatient consultations, laboratory and pharmaceutical claims, and accident compensation claims. Epidemiological methods will be used to calculate prevalences for Maori and non-Maori, by age-group, gender and socioeconomic deprivation (New Zealand Deprivation Index) where possible. Rates of 'potentially avoidable' hospitalisations and mortality as well as 'excess or under' utilisation of healthcare will be calculated as the difference between the actual rate and that expected if Maori were to have the same rates as non-Maori. A prevalence-based cost-of-illness approach will be used to estimate health inequities and the costs associated with treatment, as well as other financial and non-financial costs (such as years of life lost) over the person's lifetime. ETHICS AND DISSEMINATION: This analysis has been approved by the University of Auckland Human Participants Research Committee (Ref: 018621). Dissemination of findings will occur via published peer-reviewed articles, presentations to academic, policy and community-based stakeholder groups and via social media.

Cumulative Prevalence of Maltreatment Among New Zealand Children, 1998-2015.

OBJECTIVES: To document, via linked administrative data, the cumulative prevalence among New Zealand children of notifications to child protective services (CPS), substantiated maltreatment cases, and out-of-home placements. METHODS: We followed all children born in New Zealand in 1998 until the end of 2015 (an overall sample of 55 443 children). We determined the cumulative frequencies of notifications, substantiated maltreatment cases (by subtype), and first entries into foster care from birth through the age of 17 years. We also decomposed CPS involvement by gender. RESULTS: We found that almost 1 in 4 children had been subject to at least 1 report to CPS at age 17 years (23.5%), and 9.7% had been a victim of substantiated abuse or neglect. We also found that 3.1% had experienced out-of-home placements by age 17 years, with boys being more affected. CONCLUSIONS: Both notifications and substantiated child maltreatment are more common in New Zealand than is generally recognized, with the incidence of notifications higher than the incidence of medicated asthma among children and the prevalence of substantiations similar to the prevalence of obesity.

Injury and Mortality Among Children Identified as at High Risk of Maltreatment.

OBJECTIVES: To determine if children identified by a predictive risk model as at "high risk" of maltreatment are also at elevated risk of injury and mortality in early childhood. METHODS: We built a model that predicted a child's risk of a substantiated finding of maltreatment by child protective services for children born in New Zealand in 2010. We assigned risk scores to the 2011 birth cohort, and flagged children as "very high risk" if they were in the top 10% of the score distribution for maltreatment. We also set a less conservative threshold for defining "high risk" and examined children in the top 20%. We then compared the incidence of injury and mortality rates between very high-risk and high-risk children and the remainder of the birth cohort. RESULTS: Children flagged at both 10% and 20% risk thresholds had much higher postneonatal mortality rates than other children (4.8 times and 4.2 times greater, respectively), as well as a greater relative risk of hospitalization (2 times higher and 1.8 times higher, respectively). CONCLUSIONS: Models that predict risk of maltreatment as defined by child protective services substantiation also identify children who are at heightened risk of injury and mortality outcomes. If deployed at birth, these models could help medical providers identify children in families who would benefit from more intensive supports.

Black-White Differences in Child Maltreatment Reports and Foster Care Placements: A Statistical Decomposition Using Linked Administrative Data.

Introduction Official statistics have confirmed that relative to their presence in the population and relative to white children, black children have consistently higher rates of contact with child protective services (CPS). We used linked administrative data and statistical decomposition techniques to generate new insights into black and white differences in child maltreatment reports and foster care placements. Methods Birth records for all children born in Allegheny County, Pennsylvania, between 2008 and 2010 were linked to administrative service records originating in multiple county data systems. Differences in rates of involvement with child protective services between black and white children by age 4 were decomposed using nonlinear regression techniques. Results Black children had rates of CPS involvement that were 3 times higher than white children. Racial differences were explained solely by parental marital status (i.e., being unmarried) and age at birth (i.e., predominantly teenage mothers). Adding other covariates did not capture any further racial differences in maltreatment reporting or foster care placement rates, they simply shifted differences already explained by marital status and age to these other variables. Discussion Racial differences in rates of maltreatment reports and foster care placements can be explained by a basic model that adjusts only for parental marital status and age at the time of birth. Increasing access to early prevention services for vulnerable families may reduce disparities in child protective service involvement. Using birth records linked to other administrative data sources provides an important means to developing population-based research.

Economic burden of disease-associated malnutrition in China.

Disease-associated malnutrition (DAM) is a well-recognized problem in many countries, but the extent of its burden on the Chinese population is unclear. This article reports the results of a burden-of-illness study on DAM in 15 diseases in China. Using data from the World Health Organization (WHO), the China Health and Nutrition Survey, and the published literature, mortality and disability-adjusted life years (DALYs) lost because of DAM were calculated; a financial value of this burden was calculated following WHO guidelines. DALYs lost annually to DAM in China varied across diseases, from a low of 2248 in malaria to a high of 1 315 276 in chronic obstructive pulmonary disease. The total burden was 6.1 million DALYs, for an economic burden of US$66 billion (Chinese ¥ 447 billion) annually. This burden is sufficiently large to warrant immediate attention from public health officials and medical providers, especially given that low-cost and effective interventions are available.

Integrating care for high-risk patients in England using the virtual ward model: lessons in the process of care integration from three case sites.

BACKGROUND: Patients at high risk of emergency hospitalisation are particularly likely to experience fragmentation in care. The virtual ward model attempts to integrate health and social care by offering multidisciplinary case management to people at high predicted risk of unplanned hospitalisation. OBJECTIVE: To describe the care practice in three virtual ward sites in England and to explore how well each site had achieved meaningful integration. METHOD: Case studies conducted in Croydon, Devon and Wandsworth during 2011-2012, consisting of semi-structured interviews, workshops, and site visits. RESULTS: Different versions of the virtual wards intervention had been implemented in each site. In Croydon, multidisciplinary care had reverted back to one-to-one case management. CONCLUSIONS: To integrate successfully, virtual ward projects should safeguard the multidisciplinary nature of the intervention, ensure the active involvement of General Practitioners, and establish feedback processes to monitor performance such as the number of professions represented at each team meeting.

Children in the public benefit system at risk of maltreatment: identification via predictive modeling.

A growing body of research links child abuse and neglect to a range of negative short- and long-term health outcomes. Determining a child's risk of maltreatment at or shortly after birth provides an opportunity for the delivery of targeted prevention services. This study presents findings from a predictive risk model (PRM) developed to estimate the likelihood of substantiated maltreatment among children enrolled in New Zealand's public benefit system. The objective was to explore the potential use of administrative data for targeting prevention and early intervention services to children and families. A data set of integrated public benefit and child protection records for children born in New Zealand between January 1, 2003, and June 1, 2006, was used to develop a risk algorithm using stepwise probit modeling. Data were analyzed in 2012. The final model included 132 variables and produced an area under the receiver operating characteristic curve of 76%. Among children in the top decile of risk, 47.8% had been substantiated for maltreatment by age 5 years. Of all children substantiated for maltreatment by age 5 years, 83% had been enrolled in the public benefit system before age 2 years. This analysis demonstrates that PRMs can be used to generate risk scores for substantiated maltreatment. Although a PRM cannot replace more-comprehensive clinical assessments of abuse and neglect risk, this approach provides a simple and cost-effective method of targeting early prevention services.

How health systems could avert 'triple fail' events that are harmful, are costly, and result in poor patient satisfaction.

Health care systems in many countries are using the "Triple Aim"--to improve patients' experience of care, to advance population health, and to lower per capita costs--as a focus for improving quality. Population strategies for addressing the Triple Aim are becoming increasingly prevalent in developed countries, but ultimately success will also require targeting specific subgroups and individuals. Certain events, which we call "Triple Fail" events, constitute a simultaneous failure to meet all three Triple Aim goals. The risk of experiencing different Triple Fail events varies widely across people. We argue that by stratifying populations according to each person's risk and anticipated response to an intervention, health systems could more effectively target different preventive interventions at particular risk strata. In this article we describe how such an approach could be planned and operationalized. Policy makers should consider using this stratified approach to reduce the incidence of Triple Fail events, thereby improving outcomes, enhancing patient experience, and lowering costs.

Multidisciplinary case management for patients at high risk of hospitalization: comparison of virtual ward models in the United kingdom, United States, and Canada.

Virtual wards are a model for delivering multidisciplinary case management to people who are at high predicted risk of unplanned acute care hospitalization. First introduced in Croydon, England, in 2006, this concept has since been adopted and adapted by health care organizations in other parts of the United Kingdom and internationally. In this article, the authors review the model of virtual wards as originally described-with its twin pillars of (1) using a predictive model to identify people who are at high risk of future emergency hospitalization, and (2) offering these individuals a period of intensive, multidisciplinary preventive care at home using the systems, staffing, and daily routines of a hospital ward. The authors then describe how virtual wards have been modified and implemented in 6 sites in the United Kingdom, United States, and Canada where they are subject to formal evaluation. Like hospital wards, virtual wards vary in terms of patient selection, ward configuration, staff composition, and ward processes. Policy makers and researchers should be aware of these differences when considering the evaluation results of studies investigating the cost-effectiveness of virtual wards.