Prediction and Privacy in Healthcare Analytics

Thesis (Master's)--University of Washington, 2016-03

Prediction of breast cancer risk based on profiling with common genetic variants

BACKGROUND: Data for multiple common susceptibility alleles for breast cancer may be combined to identify women at different levels of breast cancer risk. Such stratification could guide preventive and screening strategies. However, empirical evidence for genetic risk stratification is lacking. METHODS: We investigated the value of using 77 breast cancer-associated single nucleotide polymorphisms (SNPs) for risk stratification, in a study of 33 673 breast cancer cases and 33 381 control women of European origin. We tested all possible pair-wise multiplicative interactions and constructed a 77-SNP polygenic risk score (PRS) for breast cancer overall and by estrogen receptor (ER) status. Absolute risks of breast cancer by PRS were derived from relative risk estimates and UK incidence and mortality rates. RESULTS: There was no strong evidence for departure from a multiplicative model for any SNP pair. Women in the highest 1% of the PRS had a three-fold increased risk of developing breast cancer compared with women in the middle quintile (odds ratio [OR] = 3.36, 95% confidence interval [CI] = 2.95 to 3.83). The ORs for ER-positive and ER-negative disease were 3.73 (95% CI = 3.24 to 4.30) and 2.80 (95% CI = 2.26 to 3.46), respectively. Lifetime risk of breast cancer for women in the lowest and highest quintiles of the PRS were 5.2% and 16.6% for a woman without family history, and 8.6% and 24.4% for a woman with a first-degree family history of breast cancer. CONCLUSIONS: The PRS stratifies breast cancer risk in women both with and without a family history of breast cancer. The observed level of risk discrimination could inform targeted screening and prevention strategies. Further discrimination may be achievable through combining the PRS with lifestyle/environmental factors, although these were not considered in this report.

Macroeconomic information, structural change, and the prediction of fiscal aggregates

Previous research on the prediction of fiscal aggregates has shown evidence that simple autoregressive models often provide better forecasts of fiscal variables than multivariate specifications. We argue that the multivariate models considered by previous studies are small-scale, probably burdened by overparameterization, and not robust to structural changes. Bayesian Vector Autoregressions (BVARs), on the other hand, allow the information contained in a large data set to be summarized efficiently, and can also allow for time variation in both the coefficients and the volatilities. In this paper we explore the performance of BVARs with constant and drifting coefficients for forecasting key fiscal variables such as government revenues, expenditures, and interest payments on the outstanding debt. We focus on both point and density forecasting, as assessments of a country’s fiscal stability and overall credit risk should typically be based on the specification of a whole probability distribution for the future state of the economy. Using data from the US and the largest European countries, we show that both the adoption of a large system and the introduction of time variation help in forecasting, with the former playing a relatively more important role in point forecasting, and the latter being more important for density forecasting.

Geostatistical Prediction of Ocean Outfall Plume Characteristics Based on an Autonomous Underwater Vehicle

Geostatistics has been successfully used to analyze and characterize the spatial variability of environmental properties. Besides giving estimated values at unsampled locations, it provides a measure of the accuracy of the estimate, which is a significant advantage over traditional methods used to assess pollution. In this work universal block kriging is novelty used to model and map the spatial distribution of salinity measurements gathered by an Autonomous Underwater Vehicle in a sea outfall monitoring campaign, with the aim of distinguishing the effluent plume from the receiving waters, characterizing its spatial variability in the vicinity of the discharge and estimating dilution. The results demonstrate that geostatistical methodology can provide good estimates of the dispersion of effluents that are very valuable in assessing the environmental impact and managing sea outfalls. Moreover, since accurate measurements of the plume’s dilution are rare, these studies might be very helpful in the future to validate dispersion models.

Strength prediction of single- and double-lap joints by standard and extended finite element modelling

The structural integrity of multi-component structures is usually determined by the strength and durability of their unions. Adhesive bonding is often chosen over welding, riveting and bolting, due to the reduction of stress concentrations, reduced weight penalty and easy manufacturing, amongst other issues. In the past decades, the Finite Element Method (FEM) has been used for the simulation and strength prediction of bonded structures, by strength of materials or fracture mechanics-based criteria. Cohesive-zone models (CZMs) have already proved to be an effective tool in modelling damage growth, surpassing a few limitations of the aforementioned techniques. Despite this fact, they still suffer from the restriction of damage growth only at predefined growth paths. The eXtended Finite Element Method (XFEM) is a recent improvement of the FEM, developed to allow the growth of discontinuities within bulk solids along an arbitrary path, by enriching degrees of freedom with special displacement functions, thus overcoming the main restriction of CZMs. These two techniques were tested to simulate adhesively bonded single- and double-lap joints. The comparative evaluation of the two methods showed their capabilities and/or limitations for this specific purpose.

Development and validation of a risk model for prediction of hazardous alcohol consumption in general practice attendees : the PredictAL study

Background: Little is known about the risk of progression to hazardous alcohol use in people currently drinking at safe limits. We aimed to develop a prediction model (predictAL) for the development of hazardous drinking in safe drinkers. Methods: A prospective cohort study of adult general practice attendees in six European countries and Chile followed up over 6 months. We recruited 10,045 attendees between April 2003 to February 2005. 6193 European and 2462 Chilean attendees recorded AUDIT scores below 8 in men and 5 in women at recruitment and were used in modelling risk. 38 risk factors were measured to construct a risk model for the development of hazardous drinking using stepwise logistic regression. The model was corrected for over fitting and tested in an external population. The main outcome was hazardous drinking defined by an AUDIT score >= 8 in men and >= 5 in women. Results: 69.0% of attendees were recruited, of whom 89.5% participated again after six months. The risk factors in the final predictAL model were sex, age, country, baseline AUDIT score, panic syndrome and lifetime alcohol problem. The predictAL model's average c-index across all six European countries was 0.839 (95% CI 0.805, 0.873). The Hedge's g effect size for the difference in log odds of predicted probability between safe drinkers in Europe who subsequently developed hazardous alcohol use and those who did not was 1.38 (95% CI 1.25, 1.51). External validation of the algorithm in Chilean safe drinkers resulted in a c-index of 0.781 (95% CI 0.717, 0.846) and Hedge's g of 0.68 (95% CI 0.57, 0.78). Conclusions: The predictAL risk model for development of hazardous consumption in safe drinkers compares favourably with risk algorithms for disorders in other medical settings and can be a useful first step in prevention of alcohol misuse.

Creation of a web application for smart park system and parking prediction study for system integration

In the last few years, we have observed an exponential increasing of the information systems, and parking information is one more example of them. The needs of obtaining reliable and updated information of parking slots availability are very important in the goal of traffic reduction. Also parking slot prediction is a new topic that has already started to be applied. San Francisco in America and Santander in Spain are examples of such projects carried out to obtain this kind of information. The aim of this thesis is the study and evaluation of methodologies for parking slot prediction and the integration in a web application, where all kind of users will be able to know the current parking status and also future status according to parking model predictions. The source of the data is ancillary in this work but it needs to be understood anyway to understand the parking behaviour. Actually, there are many modelling techniques used for this purpose such as time series analysis, decision trees, neural networks and clustering. In this work, the author explains the best techniques at this work, analyzes the result and points out the advantages and disadvantages of each one. The model will learn the periodic and seasonal patterns of the parking status behaviour, and with this knowledge it can predict future status values given a date. The data used comes from the Smart Park Ontinyent and it is about parking occupancy status together with timestamps and it is stored in a database. After data acquisition, data analysis and pre-processing was needed for model implementations. The first test done was with the boosting ensemble classifier, employed over a set of decision trees, created with C5.0 algorithm from a set of training samples, to assign a prediction value to each object. In addition to the predictions, this work has got measurements error that indicates the reliability of the outcome predictions being correct. The second test was done using the function fitting seasonal exponential smoothing tbats model. Finally as the last test, it has been tried a model that is actually a combination of the previous two models, just to see the result of this combination. The results were quite good for all of them, having error averages of 6.2, 6.6 and 5.4 in vacancies predictions for the three models respectively. This means from a parking of 47 places a 10% average error in parking slot predictions. This result could be even better with longer data available. In order to make this kind of information visible and reachable from everyone having a device with internet connection, a web application was made for this purpose. Beside the data displaying, this application also offers different functions to improve the task of searching for parking. The new functions, apart from parking prediction, were: - Park distances from user location. It provides all the distances to user current location to the different parks in the city. - Geocoding. The service for matching a literal description or an address to a concrete location. - Geolocation. The service for positioning the user. - Parking list panel. This is not a service neither a function, is just a better visualization and better handling of the information.

Credit default swap (CDS) prediction model & trading strategy

This project focuses on the study of different explanatory models for the behavior of CDS security, such as Fixed-Effect Model, GLS Random-Effect Model, Pooled OLS and Quantile Regression Model. After determining the best fitness model, trading strategies with long and short positions in CDS have been developed. Due to some specifications of CDS, I conclude that the quantile regression is the most efficient model to estimate the data. The P&L and Sharpe Ratio of the strategy are analyzed using a backtesting analogy, where I conclude that, mainly for non-financial companies, the model allows traders to take advantage of and profit from arbitrages.

Risk prediction of prevalent diabetes in a Swiss population using a weighted genetic score--the CoLaus Study.

AIMS/HYPOTHESIS: Several susceptibility genes for type 2 diabetes have been discovered recently. Individually, these genes increase the disease risk only minimally. The goals of the present study were to determine, at the population level, the risk of diabetes in individuals who carry risk alleles within several susceptibility genes for the disease and the added value of this genetic information over the clinical predictors. METHODS: We constructed an additive genetic score using the most replicated single-nucleotide polymorphisms (SNPs) within 15 type 2 diabetes-susceptibility genes, weighting each SNP with its reported effect. We tested this score in the extensively phenotyped population-based cross-sectional CoLaus Study in Lausanne, Switzerland (n = 5,360), involving 356 diabetic individuals. RESULTS: The clinical predictors of prevalent diabetes were age, BMI, family history of diabetes, WHR, and triacylglycerol/HDL-cholesterol ratio. After adjustment for these variables, the risk of diabetes was 2.7 (95% CI 1.8-4.0, p = 0.000006) for individuals with a genetic score within the top quintile, compared with the bottom quintile. Adding the genetic score to the clinical covariates improved the area under the receiver operating characteristic curve slightly (from 0.86 to 0.87), yet significantly (p = 0.002). BMI was similar in these two extreme quintiles. CONCLUSIONS/INTERPRETATION: In this population, a simple weighted 15 SNP-based genetic score provides additional information over clinical predictors of prevalent diabetes. At this stage, however, the clinical benefit of this genetic information is limited.

Prediction of multiple infections after severe burn trauma: a prospective cohort study.

OBJECTIVE: To develop predictive models for early triage of burn patients based on hypersusceptibility to repeated infections. BACKGROUND: Infection remains a major cause of mortality and morbidity after severe trauma, demanding new strategies to combat infections. Models for infection prediction are lacking. METHODS: Secondary analysis of 459 burn patients (≥16 years old) with 20% or more total body surface area burns recruited from 6 US burn centers. We compared blood transcriptomes with a 180-hour cutoff on the injury-to-transcriptome interval of 47 patients (≤1 infection episode) to those of 66 hypersusceptible patients [multiple (≥2) infection episodes (MIE)]. We used LASSO regression to select biomarkers and multivariate logistic regression to built models, accuracy of which were assessed by area under receiver operating characteristic curve (AUROC) and cross-validation. RESULTS: Three predictive models were developed using covariates of (1) clinical characteristics; (2) expression profiles of 14 genomic probes; (3) combining (1) and (2). The genomic and clinical models were highly predictive of MIE status [AUROCGenomic = 0.946 (95% CI: 0.906-0.986); AUROCClinical = 0.864 (CI: 0.794-0.933); AUROCGenomic/AUROCClinical P = 0.044]. Combined model has an increased AUROCCombined of 0.967 (CI: 0.940-0.993) compared with the individual models (AUROCCombined/AUROCClinical P = 0.0069). Hypersusceptible patients show early alterations in immune-related signaling pathways, epigenetic modulation, and chromatin remodeling. CONCLUSIONS: Early triage of burn patients more susceptible to infections can be made using clinical characteristics and/or genomic signatures. Genomic signature suggests new insights into the pathophysiology of hypersusceptibility to infection may lead to novel potential therapeutic or prophylactic targets.

Predicting recurrence after unprovoked venous thromboembolism: prospective validation of the updated Vienna Prediction Model.

The updated Vienna Prediction Model for estimating recurrence risk after an unprovoked venous thromboembolism (VTE) has been developed to identify individuals at low risk for VTE recurrence in whom anticoagulation (AC) therapy may be stopped after 3 months. We externally validated the accuracy of the model to predict recurrent VTE in a prospective multicenter cohort of 156 patients aged ≥65 years with acute symptomatic unprovoked VTE who had received 3 to 12 months of AC. Patients with a predicted 12-month risk within the lowest quartile based on the updated Vienna Prediction Model were classified as low risk. The risk of recurrent VTE did not differ between low- vs higher-risk patients at 12 months (13% vs 10%; P = .77) and 24 months (15% vs 17%; P = 1.0). The area under the receiver operating characteristic curve for predicting VTE recurrence was 0.39 (95% confidence interval [CI], 0.25-0.52) at 12 months and 0.43 (95% CI, 0.31-0.54) at 24 months. In conclusion, in elderly patients with unprovoked VTE who have stopped AC, the updated Vienna Prediction Model does not discriminate between patients who develop recurrent VTE and those who do not. This study was registered at www.clinicaltrials.gov as #NCT00973596.

A Meta-Analysis of Trabecular Bone Score in Fracture Risk Prediction and Its Relationship to FRAX.

Trabecular bone score (TBS) is a gray-level textural index of bone microarchitecture derived from lumbar spine dual-energy X-ray absorptiometry (DXA) images. TBS is a bone mineral density (BMD)-independent predictor of fracture risk. The objective of this meta-analysis was to determine whether TBS predicted fracture risk independently of FRAX probability and to examine their combined performance by adjusting the FRAX probability for TBS. We utilized individual-level data from 17,809 men and women in 14 prospective population-based cohorts. Baseline evaluation included TBS and the FRAX risk variables, and outcomes during follow-up (mean 6.7 years) comprised major osteoporotic fractures. The association between TBS, FRAX probabilities, and the risk of fracture was examined using an extension of the Poisson regression model in each cohort and for each sex and expressed as the gradient of risk (GR; hazard ratio per 1 SD change in risk variable in direction of increased risk). FRAX probabilities were adjusted for TBS using an adjustment factor derived from an independent cohort (the Manitoba Bone Density Cohort). Overall, the GR of TBS for major osteoporotic fracture was 1.44 (95% confidence interval [CI] 1.35-1.53) when adjusted for age and time since baseline and was similar in men and women (p > 0.10). When additionally adjusted for FRAX 10-year probability of major osteoporotic fracture, TBS remained a significant, independent predictor for fracture (GR = 1.32, 95% CI 1.24-1.41). The adjustment of FRAX probability for TBS resulted in a small increase in the GR (1.76, 95% CI 1.65-1.87 versus 1.70, 95% CI 1.60-1.81). A smaller change in GR for hip fracture was observed (FRAX hip fracture probability GR 2.25 vs. 2.22). TBS is a significant predictor of fracture risk independently of FRAX. The findings support the use of TBS as a potential adjustment for FRAX probability, though the impact of the adjustment remains to be determined in the context of clinical assessment guidelines. © 2015 American Society for Bone and Mineral Research.