Machine learning is increasingly used to predict healthcare outcomes, including cost, utilization, and quality.
We provide a high-level overview of machine learning for healthcare outcomes ...researchers and decision makers.
We introduce key concepts for understanding the application of machine learning methods to healthcare outcomes research. We first describe current standards to rigorously learn an estimator, which is an algorithm developed through machine learning to predict a particular outcome. We include steps for data preparation, estimator family selection, parameter learning, regularization, and evaluation. We then compare 3 of the most common machine learning methods: (1) decision tree methods that can be useful for identifying how different subpopulations experience different risks for an outcome; (2) deep learning methods that can identify complex nonlinear patterns or interactions between variables predictive of an outcome; and (3) ensemble methods that can improve predictive performance by combining multiple machine learning methods.
We demonstrate the application of common machine methods to a simulated insurance claims dataset. We specifically include statistical code in R and Python for the development and evaluation of estimators for predicting which patients are at heightened risk for hospitalization from ambulatory care-sensitive conditions.
Outcomes researchers should be aware of key standards for rigorously evaluating an estimator developed through machine learning approaches. Although multiple methods use machine learning concepts, different approaches are best suited for different research problems.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
In anticipation of the 2012 World Health Report, this paper was commissioned to help contextualize and critically reflect on the theme of "no health without research."
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
We estimated the relationship between soft drink consumption and obesity and diabetes worldwide.
We used multivariate linear regression to estimate the association between soft drink consumption and ...overweight, obesity, and diabetes prevalence in 75 countries, controlling for other foods (cereals, meats, fruits and vegetables, oils, and total calories), income, urbanization, and aging. Data were obtained from the Euromonitor Global Market Information Database, the World Health Organization, and the International Diabetes Federation. Bottled water consumption, which increased with per-capita income in parallel to soft drink consumption, served as a natural control group.
Soft drink consumption increased globally from 9.5 gallons per person per year in 1997 to 11.4 gallons in 2010. A 1% rise in soft drink consumption was associated with an additional 4.8 overweight adults per 100 (adjusted B; 95% confidence interval CI = 3.1, 6.5), 2.3 obese adults per 100 (95% CI = 1.1, 3.5), and 0.3 adults with diabetes per 100 (95% CI = 0.1, 0.8). These findings remained robust in low- and middle-income countries.
Soft drink consumption is significantly linked to overweight, obesity, and diabetes worldwide, including in low- and middle-income countries.
Full text
Available for:
CEKLJ, DOBA, FSPLJ, IZUM, KILJ, NUK, ODKLJ, OILJ, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK, VSZLJ
Summary Background There is widespread concern that the present economic crisis, particularly its effect on unemployment, will adversely affect population health. We investigated how economic changes ...have affected mortality rates over the past three decades and identified how governments might reduce adverse effects. Methods We used multivariate regression, correcting for population ageing, past mortality and employment trends, and country-specific differences in health-care infrastructure, to examine associations between changes in employment and mortality, and how associations were modified by different types of government expenditure for 26 European Union (EU) countries between 1970 and 2007. Findings We noted that every 1% increase in unemployment was associated with a 0·79% rise in suicides at ages younger than 65 years (95% CI 0·16–1·42; 60–550 potential excess deaths mean 310 EU-wide), although the effect size was non-significant at all ages (0·49%, −0·04 to 1·02), and with a 0·79% rise in homicides (95% CI 0·06–1·52; 3–80 potential excess deaths mean 40 EU-wide). By contrast, road-traffic deaths decreased by 1·39% (0·64–2·14; 290–980 potential fewer deaths mean 630 EU-wide). A more than 3% increase in unemployment had a greater effect on suicides at ages younger than 65 years (4·45%, 95% CI 0·65–8·24; 250–3220 potential excess deaths mean 1740 EU-wide) and deaths from alcohol abuse (28·0%, 12·30–43·70; 1550–5490 potential excess deaths mean 3500 EU-wide). We noted no consistent evidence across the EU that all-cause mortality rates increased when unemployment rose, although populations varied substantially in how sensitive mortality was to economic crises, depending partly on differences in social protection. Every US$10 per person increased investment in active labour market programmes reduced the effect of unemployment on suicides by 0·038% (95% CI −0·004 to −0·071). Interpretation Rises in unemployment are associated with significant short-term increases in premature deaths from intentional violence, while reducing traffic fatalities. Active labour market programmes that keep and reintegrate workers in jobs could mitigate some adverse health effects of economic downturns. Funding Centre for Crime and Justice Studies, King's College, London, UK; and Wates Foundation (UK).
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK
In an article that forms part of the
PLoS Medicine
series on Big Food, David Stuckler and colleagues report that unhealthy packaged foods are being consumed rapidly in low- and middle-income ...countries, consistent with rapid expansion of multinational food companies into emerging markets and fueling obesity and chronic disease epidemics.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Private sector healthcare delivery in low- and middle-income countries is sometimes argued to be more efficient, accountable, and sustainable than public sector delivery. Conversely, the public ...sector is often regarded as providing more equitable and evidence-based care. We performed a systematic review of research studies investigating the performance of private and public sector delivery in low- and middle-income countries.
Peer-reviewed studies including case studies, meta-analyses, reviews, and case-control analyses, as well as reports published by non-governmental organizations and international agencies, were systematically collected through large database searches, filtered through methodological inclusion criteria, and organized into six World Health Organization health system themes: accessibility and responsiveness; quality; outcomes; accountability, transparency, and regulation; fairness and equity; and efficiency. Of 1,178 potentially relevant unique citations, data were obtained from 102 articles describing studies conducted in low- and middle-income countries. Comparative cohort and cross-sectional studies suggested that providers in the private sector more frequently violated medical standards of practice and had poorer patient outcomes, but had greater reported timeliness and hospitality to patients. Reported efficiency tended to be lower in the private than in the public sector, resulting in part from perverse incentives for unnecessary testing and treatment. Public sector services experienced more limited availability of equipment, medications, and trained healthcare workers. When the definition of "private sector" included unlicensed and uncertified providers such as drug shop owners, most patients appeared to access care in the private sector; however, when unlicensed healthcare providers were excluded from the analysis, the majority of people accessed public sector care. "Competitive dynamics" for funding appeared between the two sectors, such that public funds and personnel were redirected to private sector development, followed by reductions in public sector service budgets and staff.
Studies evaluated in this systematic review do not support the claim that the private sector is usually more efficient, accountable, or medically effective than the public sector; however, the public sector appears frequently to lack timeliness and hospitality towards patients.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Large-scale public policy changes are often recommended to improve public health. Despite varying widely-from tobacco taxes to poverty-relief programs-such policies present a common dilemma to public ...health researchers: how to evaluate their health effects when randomized controlled trials are not possible. Here, we review the state of knowledge and experience of public health researchers who rigorously evaluate the health consequences of large-scale public policy changes. We organize our discussion by detailing approaches to address three common challenges of conducting policy evaluations: distinguishing a policy effect from time trends in health outcomes or preexisting differences between policy-affected and -unaffected communities (using difference-in-differences approaches); constructing a comparison population when a policy affects a population for whom a well-matched comparator is not immediately available (using propensity score or synthetic control approaches); and addressing unobserved confounders by utilizing quasi-random variations in policy exposure (using regression discontinuity, instrumental variables, or near-far matching approaches).
BACKGROUND:Social determinants of health (SDH) at the area level are understood to influence the likelihood of having poor glycemic control for patients with type 2 diabetes mellitus (T2DM).
...OBJECTIVES:To develop a model for predicting whether a person with T2DM has uncontrolled diabetes (hemoglobin A1c ≥9%), incorporating individual and area-level (census tract) covariates.
RESEARCH DESIGN:Development and validation of machine learning models.
SUBJECTS:Total of N=1,015,808 privately insured persons in claims data with T2DM.
MEASURES:C-statistic, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy.
RESULTS:A standard logistic regression model selecting among the available individual-level covariates and area-level SDH covariates (at the census tract level) performed poorly, with a C-statistic of 0.685, sensitivity of 25.6%, specificity of 90.1%, positive predictive value of 56.9%, negative predictive value of 70.4%, and accuracy of 68.4% on a 25% held-out validation subset of the data. By contrast, machine learning models improved upon risk prediction, with the highest performance from a random forest algorithm with a C-statistic of 0.928, sensitivity of 68.5%, specificity of 94.6%, positive predictive value of 69.8%, negative predictive value of 94.3%, and accuracy of 90.6%. SDH variables alone explained 16.9% of variation in uncontrolled diabetes.
CONCLUSIONS:A predictive model developed through a machine learning approach may assist health care organizations to identify which area-level SDH data to monitor for prediction of diabetes control, for potential use in risk-adjustment and targeting.
Sanjay Basu and colleagues explain how models are increasingly used to inform public health policy yet readers may struggle to evaluate the quality of models. All models require simplifying ...assumptions, and there are tradeoffs between creating models that are more "realistic" versus those that are grounded in more solid data. Indeed, complex models are not necessarily more accurate or reliable simply because they can more easily fit real-world data than simpler models can. Please see later in the article for the Editors' Summary.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Summary Background Official projections of the cholera epidemic in Haiti have not incorporated existing disease trends or patterns of transmission, and proposed interventions have been debated ...without comparative estimates of their effect. We used a mathematical model of the epidemic to provide projections of future morbidity and mortality, and to produce comparative estimates of the effects of proposed interventions. Methods We designed mathematical models of cholera transmission based on existing models and fitted them to incidence data reported in Haiti for each province from Oct 31, 2010, to Jan 24, 2011. We then simulated future epidemic trajectories from March 1 to Nov 30, 2011, to estimate the effect of clean water, vaccination, and enhanced antibiotic distribution programmes. Findings We project 779 000 cases of cholera in Haiti (95% CI 599 000–914 000) and 11 100 deaths (7300–17 400) between March 1 and Nov 30, 2011. We expect that a 1% per week reduction in consumption of contaminated water would avert 105 000 cases (88 000–116 000) and 1500 deaths (1100–2300). We predict that the vaccination of 10% of the population, from March 1, will avert 63 000 cases (48 000–78 000) and 900 deaths (600–1500). The proposed extension of the use of antibiotics to all patients with severe dehydration and half of patients with moderate dehydration is expected to avert 9000 cases (8000–10 000) and 1300 deaths (900–2000). Interpretation A decline in cholera prevalence in early 2011 is part of the natural course of the epidemic, and should not be interpreted as indicative of successful intervention. Substantially more cases of cholera are expected than official estimates used for resource allocation. Combined, clean water provision, vaccination, and expanded access to antibiotics might avert thousands of deaths. Funding National Institutes of Health.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK