Justin Lessler

Incubation periods of acute respiratory viral infections: a systematic review

Summary Knowledge of the incubation period is essential in the investigation and control of infectious disease, but statements of incubation period are often poorly referenced, inconsistent, or based on limited data. In a systematic review of the literature on nine respiratory viral infections of public-health importance, we identified 436 articles with statements of incubation period and 38 with data for pooled analysis. We fitted a log-normal distribution to pooled data and found the median incubation period to be 5·6 days (95% CI 4·8–6·3) for adenovirus, 3·2 days (95% CI 2·8–3·7) for human coronavirus, 4·0 days (95% CI 3·6–4·4) for severe acute respiratory syndrome coronavirus, 1·4 days (95% CI 1·3–1·5) for influenza A, 0·6 days (95% CI 0·5–0·6) for influenza B, 12·5 days (95% CI 11·8–13·3) for measles, 2·6 days (95% CI 2·1–3·1) for parainfluenza, 4·4 days (95% CI 3·9–4·9) for respiratory syncytial virus, and 1·9 days (95% CI 1·4–2·4) for rhinovirus. When using the incubation period, it is important to consider its full distribution: the right tail for quarantine policy, the central regions for likely times and sources of infection, and the full distribution for models used in pandemic planning. Our estimates combine published data to give the detail necessary for these and other applications.

Improving propensity score weighting using machine learning

Machine learning techniques such as classification and regression trees (CART) have been suggested as promising alternatives to logistic regression for the estimation of propensity scores. The authors examined the performance of various CART-based propensity score models using simulated data. Hypothetical studies of varying sample sizes (n=500, 1000, 2000) with a binary exposure, continuous outcome, and 10 covariates were simulated under seven scenarios differing by degree of non-linear and non-additive associations between covariates and the exposure. Propensity score weights were estimated using logistic regression (all main effects), CART, pruned CART, and the ensemble methods of bagged CART, random forests, and boosted CART. Performance metrics included covariate balance, standard error, per cent absolute bias, and 95 per cent confidence interval (CI) coverage. All methods displayed generally acceptable performance under conditions of either non-linearity or non-additivity alone. However, under conditions of both moderate non-additivity and moderate non-linearity, logistic regression had subpar performance, whereas ensemble methods provided substantially better bias reduction and more consistent 95 per cent CI coverage. The results suggest that ensemble methods, especially boosted CART, may be useful for propensity score weighting.

Outbreak of 2009 Pandemic Influenza A (H1N1) at a New York City School

BACKGROUND In April 2009, an outbreak of novel swine-origin influenza A (2009 H1N1 influenza) occurred at a high school in Queens, New York. We describe the outbreak and characterize the clinical and epidemiologic aspects of this novel virus. METHODS The New York City Department of Health and Mental Hygiene characterized the outbreak through laboratory confirmation of the presence of the 2009 H1N1 virus in nasopharyngeal and oropharyngeal specimens and through information obtained from an online survey. Detailed information on exposure and the onset of symptoms was used to estimate the incubation period, generation time, and within-school reproductive number associated with 2009 H1N1 influenza, with the use of established techniques. RESULTS From April 24 through May 8, infection with the 2009 H1N1 virus was confirmed in 124 high-school students and employees. In responses to the online questionnaire, more than 800 students and employees (35% of student respondents and 10% of employee respondents) reported having an influenza-like illness during this period. No persons with confirmed 2009 H1N1 influenza or with influenza-like illness had severe symptoms. A linkage with travel to Mexico was identified. The estimated median incubation period for confirmed 2009 H1N1 influenza was 1.4 days (95% confidence interval [CI], 1.0 to 1.8), with symptoms developing in 95% of cases by 2.2 days (95% CI, 1.7 to 2.6). The estimated median generation time was 2.7 days (95% CI, 2.0 to 3.5). We estimate that the within-school reproductive number was 3.3. CONCLUSIONS The findings from this investigation suggest that 2009 H1N1 influenza in the high school was widespread but did not cause severe illness. The reasons for the rapid and extensive spread of influenza-like illnesses are unknown. The natural history and transmission of the 2009 H1N1 influenza virus appear to be similar to those of previously observed circulating pandemic and interpandemic influenza viruses.

Association of Race and Age With Survival Among Patients Undergoing Dialysis

CONTEXT Many studies have reported that black individuals undergoing dialysis survive longer than those who are white. This observation is paradoxical given racial disparities in access to and quality of care, and is inconsistent with observed lower survival among black patients with chronic kidney disease. We hypothesized that age and the competing risk of transplantation modify survival differences by race. OBJECTIVE To estimate death among dialysis patients by race, accounting for age as an effect modifier and kidney transplantation as a competing risk. DESIGN, SETTING, AND PARTICIPANTS An observational cohort study of 1,330,007 incident end-stage renal disease patients as captured in the United States Renal Data System between January 1, 1995, and September 28, 2009 (median potential follow-up time, 6.7 years; range, 1 day-14.8 years). Multivariate age-stratified Cox proportional hazards and competing risk models were constructed to examine death in patients who receive dialysis. MAIN OUTCOME MEASURES Death in black vs white patients who receive dialysis. RESULTS Similar to previous studies, black patients undergoing dialysis had a lower death rate compared with white patients (232,361 deaths [57.1% mortality] vs 585,792 deaths [63.5% mortality], respectively; adjusted hazard ratio [aHR], 0.84; 95% confidence interval [CI], 0.83-0.84; P <.001). However, when stratifying by age and treating kidney transplantation as a competing risk, black patients had significantly higher mortality than their white counterparts at ages 18 to 30 years (27.6% mortality vs 14.2%; aHR, 1.93; 95% CI, 1.84-2.03), 31 to 40 years (37.4% mortality vs 26.8%; aHR, 1.46; 95% CI, 1.41-1.50), and 41 to 50 years (44.8% mortality vs 38.0%; aHR, 1.12; 95% CI, 1.10-1.14; P <.001 for interaction terms between race and each aforementioned age category), as opposed to patients aged 51 to 60 years (51.5% vs 50.9%; aHR, 0.93; 95% CI, 0.92-0.94), 61 to 70 years (64.9% vs 67.2%; aHR, 0.87; 95% CI, 0.86-0.88), 71 to 80 years (76.1% vs 79.7%; aHR, 0.85; 95% CI, 0.84-0.86), and older than 80 years (82.4% vs 83.6%; aHR, 0.87; 95% CI, 0.85-0.88). CONCLUSIONS Overall, among dialysis patients in the United States, there was a lower risk of death for black patients compared with their white counterparts. However, the commonly cited survival advantage for black dialysis patients applies only to older adults, and those younger than 50 years have a higher risk of death.

Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression.

OBJECTIVE Propensity scores for the analysis of observational data are typically estimated using logistic regression. Our objective in this review was to assess machine learning alternatives to logistic regression, which may accomplish the same goals but with fewer assumptions or greater accuracy. STUDY DESIGN AND SETTING We identified alternative methods for propensity score estimation and/or classification from the public health, biostatistics, discrete mathematics, and computer science literature, and evaluated these algorithms for applicability to the problem of propensity score estimation, potential advantages over logistic regression, and ease of use. RESULTS We identified four techniques as alternatives to logistic regression: neural networks, support vector machines, decision trees (classification and regression trees [CART]), and meta-classifiers (in particular, boosting). CONCLUSION Although the assumptions of logistic regression are well understood, those assumptions are frequently ignored. All four alternatives have advantages and disadvantages compared with logistic regression. Boosting (meta-classifiers) and, to a lesser extent, decision trees (particularly CART), appear to be most promising for use in the context of propensity score analysis, but extensive simulation studies are needed to establish their utility in practice.

Assessing the global threat from Zika virus

Global spread of Zika virus Zika virus was identified in Uganda in 1947; since then, it has enveloped the tropics, causing disease of varying severity. Lessler et al. review the historical literature to remind us that Zikas neurotropism was observed in mice even before clinical case reports in Nigeria in 1953. What determines the clinical manifestations; how local conditions, vectors, genetics, and wild hosts affect transmission and geographical spread; what the best control strategy is; and how to develop effective drugs, vaccines, and diagnostics are all critical questions that are begging for data. Science, this issue p. 663 Assessing the global threat from Zika virus. BACKGROUND First discovered in 1947, Zika virus (ZIKV) received little attention until a surge in microcephaly cases was reported after a 2015 outbreak in Brazil. The size of the outbreak and the severity of associated birth defects prompted the World Health Organization (WHO) to declare a Public Health Emergency of International Concern on 1 February 2016. In response, there has been an explosion in research and planning as the global health community has turned its attention to understanding and controlling ZIKV. Still, much of the information needed to evaluate the global health threat from ZIKV is lacking. The global threat posed by any emerging pathogen depends on its epidemiology, its clinical features, and our ability to implement effective control measures. Whether introductions of ZIKV result in epidemics depends on local ecology, population immunity, regional demographics, and, to no small degree, random chance. The same factors determine whether the virus will establish itself as an endemic disease. The burden of ZIKV spread on human health is mediated by its natural history and pathogenesis, particularly during pregnancy, and our ability to control the virus’s spread. In this Review, we examine the empirical evidence for a global threat from ZIKV through the lens of these processes, examining historic and current evidence, as well as parallel processes in closely related viruses. ADVANCES Because ZIKV was not recognized as an important disease in humans until recently, it was little studied before the recent crisis. Nevertheless, the limited data from the decades following its discovery provide important clues into ZIKV’s epidemiology and suggest that some populations were at risk for the virus for years in the mid-20th century, although this risk may predominantly have been the result of spillover infections from a sylvatic reservoir. Recent outbreaks on Yap Island (2007) and in French Polynesia (2014) provide the only previous observations of large epidemics and are the basis for the little that we do know about ZIKV’s acute symptoms (e.g., rash, fever, conjunctivitis, and arthralgia), the risk of birth defects, such as microcephaly (estimated to be 1 per 100 in French Polynesia), and the incidence of severe neurological outcomes (e.g., Guillain-Barré is estimated to occur in approximately 2 out of every 10,000 cases). The observation of an association between ZIKV and a surge in microcephaly cases in Brazil and the subsequent declaration of a Public Health Emergency of International Concern by the WHO have rapidly accelerated research into the virus. Small, but very important, studies have begun to identify the substantial risk the virus can pose throughout a pregnancy, and careful surveillance has established that ZIKV can be transmitted sexually. Numerous modeling studies have helped to estimate the potential range of ZIKV and measured its reproductive number R0 (estimates range from 1.4 to 6.6), a key measure of transmissibility in a number of settings. Still, it remains unclear whether the recent epidemic in the Americas is the result of fundamental changes in the virus or merely a chance event. OUTLOOK ZIKV research is progressing rapidly, and over the coming months and years our understanding of the virus will undoubtedly deepen considerably. Key questions about the virus’s range, its ability to persist, and its clinical severity will be answered as the current epidemic in the Americas runs its course. Moving forward, it is important that information on ZIKV be placed within the context of its effect on human health and that we remain cognizant of the structure of postinvasion epidemic dynamics as we respond to this emerging threat. The effect of ZIKV is a function of the local transmission regime and viral pathogenesis. (A) Many countries cannot maintain ongoing vector-mediated ZIKV transmission and are only at risk from importation by travelers and limited onward transmission (e.g., through sex). (B) If conditions are appropriate, importations can lead to postinvasion epidemics with high incidence across age ranges, after which the virus may go locally extinct or remain endemic

The impact of the demographic transition on dengue in Thailand: insights from a statistical analysis and mathematical modeling.

Analyzing data from Thailands 72 provinces, Derek Cummings and colleagues find that decreases in birth and death rates can explain the shift in age distribution of dengue hemorrhagic fever.

Modeling infectious disease dynamics in the complex landscape of global health

Mathematical modeling of infectious diseases The spread of infectious diseases can be unpredictable. With the emergence of antibiotic resistance and worrying new viruses, and with ambitious plans for global eradication of polio and the elimination of malaria, the stakes have never been higher. Anticipation and measurement of the multiple factors involved in infectious disease can be greatly assisted by mathematical methods. In particular, modeling techniques can help to compensate for imperfect knowledge, gathered from large populations and under difficult prevailing circumstances. Heesterbeek et al. review the development of mathematical models used in epidemiology and how these can be harnessed to develop successful control strategies and inform public health policy. Science, this issue 10.1126/science.aaa4339 BACKGROUND Despite many notable successes in prevention and control, infectious diseases remain an enormous threat to human and animal health. The ecological and evolutionary dynamics of pathogens play out on a wide range of interconnected temporal, organizational, and spatial scales that span hours to months, cells to ecosystems, and local to global spread. Some pathogens are directly transmitted between individuals of a single species, whereas others circulate among multiple hosts, need arthropod vectors, or persist in environmental reservoirs. Many factors, including increasing antimicrobial resistance, human connectivity, population growth, urbanization, environmental and land-use change, as well as changing human behavior, present global challenges for prevention and control. Faced with this complexity, mathematical models offer valuable tools for understanding epidemiological patterns and for developing and evaluating evidence for decision-making in global health. ADVANCES During the past 50 years, the study of infectious disease dynamics has matured into a rich interdisciplinary field at the intersection of mathematics, epidemiology, ecology, evolutionary biology, immunology, sociology, and public health. The practical challenges range from establishing appropriate data collection to managing increasingly large volumes of information. The theoretical challenges require fundamental study of many-layered, nonlinear systems in which infections evolve and spread and where key events can be governed by unpredictable pathogen biology or human behavior. In this Review, we start with an examination of real-time outbreak response using the West African Ebola epidemic as an example. Here, the challenges range from underreporting of cases and deaths, and missing information on the impact of control measures to understanding human responses. The possibility of future zoonoses tests our ability to detect anomalous outbreaks and to estimate human-to-human transmissibility against a backdrop of ongoing zoonotic spillover while also assessing the risk of more dangerous strains evolving. Increased understanding of the dynamics of infections in food webs and ecosystems where host and nonhost species interact is key. Simultaneous multispecies infections are increasingly recognized as a notable public health burden, yet our understanding of how different species of pathogens interact within hosts is rudimentary. Pathogen genomics has become an essential tool for drawing inferences about evolution and transmission and, here but also in general, heterogeneity is the major challenge. Methods that depart from simplistic assumptions about random mixing are yielding new insights into the dynamics of transmission and control. There is rapid growth in estimation of model parameters from mismatched or incomplete data, and in contrasting model output with real-world observations. New data streams on social connectivity and behavior are being used, and combining data collected from very different sources and scales presents important challenges. All these mathematical endeavors have the potential to feed into public health policy and, indeed, an increasingly wide range of models is being used to support infectious disease control, elimination, and eradication efforts. OUTLOOK Mathematical modeling has the potential to probe the apparently intractable complexity of infectious disease dynamics. Coupled to continuous dialogue between decision-makers and the multidisciplinary infectious disease community, and by drawing on new data streams, mathematical models can lay bare mechanisms of transmission and indicate new approaches to prevention and control that help to shape national and international public health policy. Modeling for public health. Policy questions define the model’s purpose. Initial model design is based on current scientific understanding and the available relevant data. Model validation and fit to disease data may require further adaptation; sensitivity and uncertainty analysis can point to requirements for collection of additional specific data. Cycles of model testing and analysis thus lead to policy advice and improved scientific understanding. Despite some notable successes in the control of infectious diseases, transmissible pathogens still pose an enormous threat to human and animal health. The ecological and evolutionary dynamics of infections play out on a wide range of interconnected temporal, organizational, and spatial scales, which span hours to months, cells to ecosystems, and local to global spread. Moreover, some pathogens are directly transmitted between individuals of a single species, whereas others circulate among multiple hosts, need arthropod vectors, or can survive in environmental reservoirs. Many factors, including increasing antimicrobial resistance, increased human connectivity and changeable human behavior, elevate prevention and control from matters of national policy to international challenge. In the face of this complexity, mathematical models offer valuable tools for synthesizing information to understand epidemiological patterns, and for developing quantitative evidence for decision-making in global health.

Interactions between serotypes of dengue highlight epidemiological impact of cross-immunity

Dengue, a mosquito-borne virus of humans, infects over 50 million people annually. Infection with any of the four dengue serotypes induces protective immunity to that serotype, but does not confer long-term protection against infection by other serotypes. The immunological interactions between serotypes are of central importance in understanding epidemiological dynamics and anticipating the impact of dengue vaccines. We analysed a 38-year time series with 12 197 serotyped dengue infections from a hospital in Bangkok, Thailand. Using novel mechanistic models to represent different hypothesized immune interactions between serotypes, we found strong evidence that infection with dengue provides substantial short-term cross-protection against other serotypes (approx. 1–3 years). This is the first quantitative evidence that short-term cross-protection exists since human experimental infection studies performed in the 1950s. These findings will impact strategies for designing dengue vaccine studies, future multi-strain modelling efforts, and our understanding of evolutionary pressures in multi-strain disease systems.

Weight Trimming and Propensity Score Weighting

Propensity score weighting is sensitive to model misspecification and outlying weights that can unduly influence results. The authors investigated whether trimming large weights downward can improve the performance of propensity score weighting and whether the benefits of trimming differ by propensity score estimation method. In a simulation study, the authors examined the performance of weight trimming following logistic regression, classification and regression trees (CART), boosted CART, and random forests to estimate propensity score weights. Results indicate that although misspecified logistic regression propensity score models yield increased bias and standard errors, weight trimming following logistic regression can improve the accuracy and precision of final parameter estimates. In contrast, weight trimming did not improve the performance of boosted CART and random forests. The performance of boosted CART and random forests without weight trimming was similar to the best performance obtainable by weight trimmed logistic regression estimated propensity scores. While trimming may be used to optimize propensity score weights estimated using logistic regression, the optimal level of trimming is difficult to determine. These results indicate that although trimming can improve inferences in some settings, in order to consistently improve the performance of propensity score weighting, analysts should focus on the procedures leading to the generation of weights (i.e., proper specification of the propensity score model) rather than relying on ad-hoc methods such as weight trimming.