Statistical Confidence Intervals Research Articles

Valid inferences about soil carbon in heterogeneous landscapes

Using soil organic carbon (SOC) to generate carbon offsets requires reliably quantifying SOC sequestration. However, accuracy of SOC measurement is limited by inherent spatial heterogeneity, variability of laboratory assays, unmet statistical assumptions, and the relatively small magnitude of SOC changes over time, among other things. Most SOC measurement protocols currently used to generate offsets for C markets do not adequately address these issues, threatening to undermine climate change mitigation efforts. Using analyses and simulations from 1,117 soil samples collected from California crop and rangelands, we quantified measurement errors and sources of uncertainty to optimize SOC measurement. We demonstrate that (1) spatial heterogeneity is a primary driver of uncertainty; (2) dry combustion assays contribute little to uncertainty, although inorganic C can increase error; (3) common statistical methods—Student’s t-test and its relatives—can be unreliable for SOC (e.g. at low to medium sample sizes or when the distribution of SOC is skewed), which can lead to incorrect interpretations of SOC sequestration; and (4) common sample sizes (10–30 cores) are insufficiently powered to detect the modest SOC changes expected from management in heterogeneous agricultural landscapes. To reduce error and improve the reliability of future SOC offsets, protocols should: (1) require power analyses that include spatial heterogeneity to determine minimum sample sizes, rather than allowing arbitrarily small sample sizes; (2) minimize the use of compositing; (3) require dry combustion analysis, by the same lab for all assays; and (4) use nonparametric statistical tests and confidence intervals to control Type I error rates. While these changes might increase costs, they will make SOC estimates more accurate and more reliable, adding credibility to soil management as a climate change mitigation strategy.

Statistical Significance, Effect Size and Confidence Intervals

Article Type: Editorial Title: Statistical Significance, Effect Size and Confidence Intervals Year: 2022; Volume: 2; Issue: 4; Page No: 3–4 Author: Karuthan Chinna https://doi.org/10.55349/ijmsnr.20222434 Affiliation: Professor, Faculty of Business and Management, UCSI University, Malaysia. Email ID: karuthan@ucsiuniversity.edu.my Phone

Spatiotemporal classification of heavy rainfall patterns to characterize hydrographs in a high-resolution ensemble climate dataset

Peak discharge in rivers mostly varies depending on spatiotemporal characteristics of the rainfall, even with the same total rainfall amount. However, the spatiotemporal patterns of heavy rainfall such as rainfall levels used for river planning are difficult to determine due to the limited number of rainfall observations. We propose a spatiotemporal classification method for massive-ensemble rainfall datasets produced through dynamical downscaling with a regional climate model to clarify the possible rainfall patterns. This classification method was applied to several hundred heavy rainfall events from the massive-ensemble climate dataset, corresponding to identical exceedance probabilities within a given statistical confidence interval (i.e., 95% confidence interval of the 150-year return period in the probability limit test). The new classification method detected spatiotemporal rainfall patterns affecting peak discharges in the main river and tributaries within the target basin. These classified rainfall patterns can be used to investigate damage scenarios in the target basin, which are difficult to determine from the observed rainfall patterns. As a result, the proposed method with the massive-ensemble climate dataset will contribute to flood control planning.

Statistical Guidance to Authors at Top-Ranked Journals across Scientific Disciplines

Scientific journals may counter the misuse, misreporting, and misinterpretation of statistics by providing guidance to authors. We described the nature and prevalence of statistical guidance at 15 journals (top-ranked by Impact Factor) in each of 22 scientific disciplines across five high-level domains (N = 330 journals). The frequency of statistical guidance varied across domains (Health & Life Sciences: 122/165 journals, 74%; Multidisciplinary: 9/15 journals, 60%; Social Sciences: 8/30 journals, 27%; Physical Sciences: 21/90 journals, 23%; Formal Sciences: 0/30 journals, 0%). In one discipline (Clinical Medicine), statistical guidance was provided by all examined journals and in two disciplines (Mathematics and Computer Science) no examined journals provided statistical guidance. Of the 160 journals providing statistical guidance, 93 had a dedicated statistics section in their author instructions. The most frequently mentioned topics were confidence intervals (90 journals) and p-values (88 journals). For six “hotly debated” topics (statistical significance, p-values, Bayesian statistics, effect sizes, confidence intervals, and sample size planning/justification) journals typically offered implicit or explicit endorsement and rarely provided opposition. The heterogeneity of statistical guidance provided by top-ranked journals within and between disciplines highlights a need for further research and debate about the role journals can play in improving statistical practice.

The impact of the SARS-CoV-2 pandemic on cause-specific- mortality: a systematic literature review

Abstract Background Although investigating the patterns of COVID-19 excess mortality (EM) is relevant, understanding the effects of the pandemic on cause-specific mortality is even crucial and should also be assessed, as this metric allows for a more detailed analysis of the true impact of the pandemic. The aim of this systematic literature review is to estimate the impact of the pandemic on different causes of death, providing a quantitative and qualitative analysis of the phenomenon. Methods We searched MEDLINE to identify studies that reported cause-specific mortality during the COVID-19 pandemic. We adopted several inclusion criteria: original article; assessed at least one cause-specific mortality during the pandemic; assessed causes of deaths using the ICD-10 classification; reporting of at least one of the following outcomes: cause-specific mortality estimates or cause-specific EM; full-length articles. Several relevant data were extracted (e.g. publication year, data stratification, territory, country income level, all-cause EM, and cause-specific mortality, etc.). Results The search identified 548 articles. After title, abstract and full-text screening, we extracted relevant data from the final set of 14 articles. Cause-specific mortality was reported using different units of measurement. Only 9 studies reported the statistical significance and/or confidence intervals. The most frequently analyzed causes of death were cardiovascular diseases (n = 11), cancer (n = 7), diabetes (n = 6), and suicide (n = 5). We found very heterogeneous patterns of cause-specific mortality, for all the specific causes of deaths, except for suicide and road accident. Conclusions The impact of the pandemic on cause-specific deaths has been very heterogeneous and the analyses conducted so far are not exhaustive. We advocate for the urgent need to find a consensus to define uniform methodological approaches to establish the true burden of the COVID-19 pandemic on non-COVID-19 mortality. Key messages • We reviewed the body of literature to estimate the impact of the COVID-19 pandemic on different causes of death, and to provide a quantitative and qualitative analysis of the phenomenon. • We did not identify unique patterns of cause-specific mortality due to too varied approaches in terms of disease classification and coding, and methodologies used for estimating mortality.

Mineral fertilization in tropical reforestation: seedling growth and survival of ten native tree species in Tabasco, Mexico

A reforestation trial was carried out with ten native tree species under four different doses of mineral fertilizer in the lowland tropics in Tabasco, Mexico. In six one-hectare plots, 2415 seedlings of Brosimum alicastrum, Calophyllum brasiliense, Ceiba pentandra, Colubrina arborescens, Diphysa americana, Enterolobium cyclocarpum, Handroanthus chrysanthus, Pimenta dioica, Pouteria sapota, and Tabebuia rosea were planted in a randomized design. The mineral fertilizer consisted of a single initial dose of 0 g, 150 g, 375 g or 750 g nitrogen-phosphorus-potassium (NPK), as well as micronutrients (Ca, Cu, Fe, Mg, Mn, Zn, S). First-year survival and median height increment of transplanted seedlings were analyzed with statistical confidence intervals. Whereas overall first-year survival for Ceiba pentandra for all fertilization doses combined was 66%, it was only 12% for Brosimum alicastrum. On the other hand, median height increment was 102 cm for Enterolobium cyclocarpum, whereas it was only 7 cm for Pimenta dioica. The survival of six of the planted species was apparently affected by the salinity effect caused by mineral fertilization. The only species that benefited from mineral fertilization in a statistically significant way was C. pentandra. In conclusion, mineral fertilization in tropical reforestation is not recommended in general. Appropriate species selection is far more important for favorable survivorship and height increment.

Statistical terrain model with geometric feature detection based on GPU using LiDAR on vehicles

Digital terrain models based on light detection and ranging (LiDAR) are widely used in ground robots to achieve autonomous navigation. Very few studies have been conducted on terrain modelling for preview information, which acts as an input for a vehicle’s active suspension system. In this study, we proposed an algorithm to construct a real-time terrain model from multi-frame LiDAR points according to the applications for preview control. The terrain model consisted of an elevation map with confidence intervals and feature detection. First, the statistical confidence interval of the elevation map represented the reliability of the mean height. Further, we used spatial signal processing techniques to detect the uneven geometric features of the terrain. The key point feature on the trajectory was detected by the Laplacian of Gaussian method. Finally, the raw point cloud data processing procedure was based on parallel multi-thread techniques combined with GPU programming, which significantly promotes computation efficiency to enable real-time implementation. Thus, the proposed terrain model integrates reliability issues and geometric features, which act as inputs for preview controllers or motion planning. Experimental results proved that our method has a competitive performance (in terms of precision and efficiency) compared with the state-of-the-art methods, and runs in real-time at 50 Hz.

The Association of Sleep Duration with Breakfast Patterns and Snack Behaviors among Chinese Children Aged 6 to 17 Years: Chinese National Nutrition and Health Surveillance 2010-2012.

A significant increase in the prevalence of short sleep among children has been observed. Short sleep may be associated with unhealthy breakfast and snacking behaviors. The purpose of the current study was to explore the associations of short sleep with breakfast and snacking behaviors among children. Data were obtained from the 2010–2012 China National Nutrition and Health Surveillance (CNNHS). A total of 5254 children aged 6 to 17 years were included. Sleep duration was classified into three categories: moderate sleep, slightly short sleep, and severely short sleep. Breakfast behaviors included skipping breakfast, food diversity, intake of energy and macronutrients, and their proportion of daily total intake. Snack behaviors included snack consumption rate/frequency, types, intake of energy and macronutrients, and proportion of daily total intake. Multiple linear regression and multivariate logistic regression were used for analysis, with models adjusted for the potential effects of gender, age, region, and family income level. The bootstrapping method was used to calculate the 95% confidence intervals of the model statistics. Results showed that slightly short sleep (OR = 1.15, 95%CI: 1.00, 1.33)) and severely short sleep (OR = 1.36, 95%CI: 1.04, 1.77) was related to higher rates of skipping breakfast compared to moderate sleep. Severely short sleep was associated with higher energy (β = 28.44, 95%CI: 31.97, 44.70), carbohydrate (β = 6.62, 95%CI: 8.29, 8.84) and protein (β = 1.17, 95%CI: 1.44, 1.70) intake at breakfast and breakfast accounted for a higher proportion of total daily energy (β = 1.39, 95%CI: 1.48, 2.52), protein (β = 2.26, 95%CI: 3.16, 5.84) and carbohydrate (β = 0.83, 95%CI: 0.07, 3.41). Severely short sleep was associated with higher energy (β = 27.4, 95%CI: 18.64, 69.41), protein (β = 0.8, 95%CI: 0.48, 2.40), and fat (β = 1.40, 95%CI: 1.21, 3.16) intake at snacks and snacks accounted for a higher proportion of total daily protein intake (β = 1.23, 95%CI: 0.71, 3.58) and fat intake (β = 2.74, 95%CI: 3.13, 6.09). Slightly short sleep was associated with higher energy (β = 7.28, 95%CI: 0.15, 28.13) and carbohydrate (β = 1.67, 95%CI: 0.86, 5.73) intake at snacks and snacks accounted for a higher proportion of total daily carbohydrate intake. Children with severely short sleep were more likely to choose sugar-sweetened beverages (SSBs) as snacks (16.5%) and intake them more frequently, at a daily consumption of 204.7 g and 26.7 g per night. Overall, short sleep was associated with unhealthy breakfast patterns and snack behaviors among children. Children with short sleep had higher intake of energy and macronutrients at breakfast and snacks compared with those with moderate sleep. Promoting adequate sleep among children may have a positive effect on developing healthy eating behaviors.

Epidemiological analysis of coronary heart disease and its main risk factors: are their associations multiplicative, additive, or interactive?

Objective The purpose of this study was to discover how considering multiplicative, additive, and interactive effects modifies results of a prospective cohort study on coronary heart disease (CHD) incidence and its main risk factors. Material and methods The Kuopio Ischaemic Heart Disease Risk Factor (KIHD) Study provided the study material, 2682 Eastern Finnish middle-aged men, followed since the 1980s. We applied multiplicative and additive survival models together with different statistical metrics and confidence intervals for risk ratios and risk differences to estimate the nature of associations. Results The mean (SD) follow-up time among men who were free of CHD at baseline (n = 1958) was 21.4 (10.4) years, and 717 (37%) of them had the disease and 301 (15%) died for CHD before the end of follow-up. All tested non-modifiable and modifiable risk factors statistically significantly predicted CHD incidence. We detected three interactions: circulating low-density lipoprotein cholesterol (LDL-C) × age, obesity × age, and obesity × smoking of which LDL-C × age was the most evident one. High LDL-C increased the risk of CHD more among men younger than 50 [risk ratio (RR) 2.10] than those older than 50 (RR 1.22). LDL-C status was the only additive covariate. The additive effect of high LDL-C increased almost linearly up to 18 years and then reached a plateau. The simple multiplicative survival model stressed glycemic status as the strongest modifiable risk factor for developing CHD [hazard ratio (HR) for diabetes vs. normoglycemia was 2.69], whereas the model considering interactions and time dependence emphasised the role of LDL-C status (HR for high LDL-C vs. lower than borderline was 4.43). Age was the strongest non-modifiable predictor. Conclusions Including covariate interactions and time dependence in survival models potentially refine results of epidemiological analyses and ease to define the order of importance across CHD risk factors. KEY MESSAGES Including covariate interactions and time dependence in survival models potentially refine results of epidemiological analyses on coronary heart disease. Including covariate interactions and time dependence in survival models potentially ease to define the order of importance across coronary heart disease risk factors.

Model-free bootstrap for a general class of stationary time series

A model-free bootstrap procedure for a general class of stationary time series is introduced. The theoretical framework is established, showing asymptotic validity of bootstrap confidence intervals for many statistics of interest. In addition, asymptotic validity of one-step ahead bootstrap prediction intervals is also demonstrated. Finite-sample experiments are conducted to empirically confirm the performance of the new method, and to compare with popular methods such as the block bootstrap and the autoregressive (AR)-sieve bootstrap.

An energy harvesting reactor for high voltage overhead insulated shield wires considering randomness of meteorological conditions

Harvesting energy from the insulated shield wire with tuning is expected to be a possible solution to the problem of high power supply for online monitoring devices on high voltage overhead transmission lines. However, the randomness of meteorological conditions, such as wind and temperature, may cause fluctuations in the energy harvesting system's power and voltage. This paper presents an integrated energy harvesting reactor with both tuning and transformer functions, which can also consider the randomness of meteorological conditions. The statistical confidence interval of equivalent circuit parameters is numerically calculated based on meteorological data collected along a transmission line for a whole year from 2016 to 2020. On this basis, a combined magnetic circuit for the energy harvesting reactor is proposed to achieve a self-adaptive negative-feedback regulation, and parameter optimization is carried out and validated by field-circuit coupled simulation. Finally, an energy harvesting reactor prototype is developed and tested. The statistical analysis results on the equivalent circuit parameters show that they are approximately subject to Gaussian distribution. Besides, it is found from the experimental results that the maximum power under rated equivalent circuit parameters is 935 W, and the maximum power can reach 800–1000 W within the fluctuation range of equivalent circuit parameters considering meteorological randomness. Overall, the experimental results of the harvested power varying with load resistance conforms to the simulation results. Furthermore, compared to using a deeply saturated tuned reactor, the energy harvesting reactor shows better robustness to the randomness of meteorological conditions and the total harmonic content of output voltage is reduced by 64%. Therefore, the proposed energy harvesting reactor for high voltage overhead insulated shield sires shows more potential for a practical application than the existing methods.

Machine Learning Outperforms Existing Methods to Predict Post-Operative Mortality in Patients Bridged to Heart Transplantation with Temporary Mechanical Circulatory Support

<h3>Purpose</h3> There is a paucity of methods to accurately predict post-transplant mortality in patients bridged to heart transplantation with temporary mechanical circulatory support (tMCS). We built a machine learning (ML) model to predict 1-year post-operative mortality in this population and compared it to an existing risk scoring method. <h3>Methods</h3> We identified adult patients in the United Network for Organ Sharing (UNOS) database undergoing heart transplantation between 2009 and 2018 who were supported with tMCS at the time of transplant. We constructed a machine learning model using extreme gradient boosting (XGB) with an 80:20 train:test split to predict 1-year post-operative mortality using variables available at or before transplantation. XGB results were compared against a logistic regression model (IMPACT-LR) constructed based on variables from the Index for Mortality Prediction After Cardiac Transplantation (IMPACT) score. Statistical comparisons and confidence intervals (CIs) were made using DeLong's test. <h3>Results</h3> A total of 1584 patients were included, with a median age of 56 (interquartile range: 46-62) and 74% male. Actual 1-year mortality was 11.3%. The area under the receiver operator characteristics curve (AUC) for the XGB model was 0.70 (95% CI: 0.62-0.78). The most important variables predictive of 1-year mortality included recipient functional status, age, pulmonary capillary wedge pressure, cardiac output, serum creatinine, ECMO usage, and donor aspartate aminotransferase. XGB significantly outperformed IMPACT-LR (p<0.001), which had an AUC of 0.51 (95% CI: 0.48-0.53) (Figure 1). <h3>Conclusion</h3> An ML model trained on a large clinical database outperformed existing methods in predicting 1-year mortality for bridge to heart transplantation with tMCS. ML may be used to enhance pre-transplant evaluation in this markedly high risk population for which existing models are not well validated.

Prevalence and Risk Factors for Acute Kidney Injury Among Adults Undergoing Cardiac Interventions in King Abdulaziz University Hospital: A Retrospective Review.

BackgroundAcute kidney injury (AKI) is a syndrome that has been receiving considerable attention as a common risk in cardiac surgeries, which has consequences for short- and long-term survival implications, even for those who do not progress to renal failure. There have been limited studies in the Middle East, and specifically in the Kingdom of Saudi Arabia (KSA). Therefore, our study aimed to identify the prevalence of and risk factors for AKIs following adult cardiac interventions during 2010-2020 at the King Abdulaziz University Hospital (KAUH), Jeddah, Saudi Arabia.MethodsSetting and Design: A retrospective medical record review was conducted among all the adult patients who underwent cardiac interventions and developed AKIs between 2010 and 2020. Google forms were used to extract the data from the hospital records. About statistical analysis binary logistic regression analyses, relative risks (RRs), and confidence intervals (CI) were used to determine the associations among the variables.ResultsAfter applying the inclusion and exclusion criteria, 564 patients were included. Their baseline demographic, clinical, biological, and operative characteristics were analyzed. AKIs developed in 110 (19.5%) patients and patients with diabetes were more likely to develop AKIs (P < 0.012, RR = 2.280, CI = 1.198-4.339). Hypertension showed a strong effect in the development of AKIs (P < 0.004, RR = 2.865, CI = 1.391-5.900). Moreover, patients who suffered from chronic heart failure were more prone to the development of AKIs (P < 0.008, RR = 4.189, CI =1.452-12.087). Furthermore, anemia with significant P-values (<0.002), and CIs of 1.509-6.822, indicated that these patients were more likely to develop AKIs (3.209 times).ConclusionWe demonstrated that AKIs are frequent complications in adults after cardiac interventions and were associated with poor outcomes. Risk factors for the development of AKIs were identified to be diabetes mellitus, hypertension, chronic heart failure, and anemia. Further investigation of this cohort is necessary to better understand the problem of kidney injuries.

Comparison between 2000 and 2018 on the reporting of statistical significance and clinical relevance in physiotherapy clinical trials in six major physiotherapy journals: a meta-research design

DesignMeta-research.ObjectiveTo compare the prevalence of reporting p values, effect estimates and clinical relevance in physiotherapy randomised controlled trials (RCTs) published in the years 2000 and 2018.MethodsWe performed a meta-research study...

Forty-year Tunisian bibliometrics of general surgery theses in the four national faculties of medicine (1980–2019)

ABSTRACT The objective of this work was to establish the bibliometric profile of Tunisian theses in ‘general surgery’ and to describe their themes, their study designs, and their writing quality. This is a retrospective descriptive bibliometric study, covering all the theses in medicine in the specialty of ‘general surgery’, defended in the four medical faculties of Tunisia, during the forty last years from 1980 to 2019. During the study period, 739 theses in ‘general surgery’ were discussed in Tunisia, with an average of 19 theses per year. The most studied research topic was emergencies (41%), followed by common surgical pathologies (26%) and digestive oncology (21.5%). Descriptive studies and case studies represented the majority of study designs with respective proportions of 56.9% and 40.6%. Only 20.7% of these theses had a scientific writing quality deemed satisfactory. The least respected elements in writing their summaries were statistical (confidence intervals and standard deviations) and documentary (keywords). Despite the plethora of themes of Tunisian theses in ‘general surgery’, their basic methodology and their editorial non conformity require the educational reform of the dissertations, both doctoral students and supervisors, by strengthening their skills in research methodology and scientific communication written.

Comparison of pediatric performance athlete injuries presenting to emergency departments

ABSTRACT Objective Pediatric injuries in performance sports represent a significant healthcare burden and account for over 50,000 annual Emergency Department (ED) visits in the United States. The objective of this study was to characterize and compare pediatric injury presentation across the most common performance sports. Methods The National Electronic Injury Surveillance System (NEISS) database was retrospectively analyzed for pediatric injuries (3–18 years) related to gymnastics, dance, or cheerleading from 2015–2019. Cases were categorized as children (˂11 years) or adolescent (≥11 years). Injuries were categorized as orthopedic (fractures, dislocations, sprain, strains), non-orthopedic (contusion, internal injury, laceration), concussion, or other/unknown. Case narratives were used to categorize mechanism of injury as contact or non-contact. Appropriate sample NEISS weights estimation was applied for statistical analysis and Confidence Intervals (CI). Results A total of 393,110 injuries were observed over the five-year study period, with a mean of 78,622 annual injuries. Gymnastics, dance, and cheerleading accounted for 136,422 injuries, 96,416 injuries, and 160,272 injuries, respectively. Most cases were adolescent (71%; 95% CI: 68–74%) and occurred in a sports-related setting (65%, 95% CI: 57–72%). Gymnastics had the highest proportion of injuries among children (50%) as compared to dancers (25%) or cheerleaders (12%) (p < 0.01). Non-contact injuries most affected the lower extremity (43–68%) and resulted in an orthopedic diagnosis (63–71%), and contact injuries had a higher proportion of injuries affecting the head, neck, and face (29–51%) and non-orthopedic diagnoses (29–38%). Gymnastics had the most upper extremity non-contact injuries (42%) and dance the most lower extremity non-contact injuries (68%) (p < 0.01). Cheerleading had the highest incidence of overall concussions (8%), contact injuries (47%), and concussions due to contact injury (15%). Conclusion Pediatric gymnasts, cheerleaders, and dancers have important similarities and differences in injury pattern which may lead to the development of sport-specific injury prevention programs for pediatric performance athletes.

An Overview of Intellectual Disability: Definition, Diagnosis, Classification, and Systems of Supports (12th ed.).

An Overview of Intellectual Disability: Definition, Diagnosis, Classification, and Systems of Supports (12th ed.).

Image segmentation by using thresholding technique in two stages

Image segmentation can be defined as a cutting or segmenting process of the digital image into many useful points which are called segmentation, that includes image elements contribute with certain attributes different form Pixel that constitute other parts. Two phases were followed in image processing by the researcher in this paper. At the beginning, pre-processing image on images was made before the segmentation process through statistical confidence intervals that can be used for estimate of unknown remarks suggested by Acho & Buenestado in 2018. Then, the second phase includes image segmentation process by using "Bernsen's Thresholding Technique" in the first phase. The researcher drew a conclusion that in case of utilizing the statistical confidence intervals beside Bernsen's Thresholding technique it can give better results in image segmentation. This method is characterized with different performance if it is compared with the regular Bernsen's thresholding technique during the direct image segmentation in both cases,namely , in case of natural image status or adding speckle noise perturbation.

Exact inference for fixed-effects meta-analysis of proportions.

Meta-analysis of proportions is conceptually simple: Faced with a binary outcome in multiple studies, we seek inference on some overall proportion of successes/failures. Under common effect models, exact inference has long been available, but is not when we more realistically allow for heterogeneity of the proportions. Instead a wide range of nonexact fixed-effects methods are used, the interpretation of some of which is challenging. In this paper, we present methods for exact statistical tests and confidence intervals for fixed-effects meta-analysis of proportions. These methods retain the interpretability of the underlying parameter of interest, and can be implemented in straightforward software. We also show how our inference on the overall proportion is compatible with exact inference on heterogeneity of proportions. An illustrative example from a recent kidney disease study shows how the method's performance can be assessed in practice.

Survival Augmented Patient Preference Incorporated Reinforcement Learning to Evaluate Tailoring Variables for Personalized Healthcare

In this paper, we consider personalized treatment decision strategies in the management of chronic diseases, such as chronic kidney disease, which typically consists of sequential and adaptive treatment decision making. We investigate a two-stage treatment setting with a survival outcome that could be right censored. This can be formulated through a dynamic treatment regime (DTR) framework, where the goal is to tailor treatment to each individual based on their own medical history in order to maximize a desirable health outcome. We develop a new method, Survival Augmented Patient Preference incorporated reinforcement Q-Learning (SAPP-Q-Learning) to decide between quality of life and survival restricted at maximal follow-up. Our method incorporates the latent patient preference into a weighted utility function that balances between quality of life and survival time, in a Q-learning model framework. We further propose a corresponding m-out-of-n Bootstrap procedure to accurately make statistical inferences and construct confidence intervals on the effects of tailoring variables, whose values can guide personalized treatment strategies.