Development and validation of predictive models for unplanned hospitalization in the Basque Country: analyzing the variability of non-deterministic algorithms.

BACKGROUND: The progressive ageing in developed countries entails an increase in multimorbidity. Population-wide predictive models for adverse health outcomes are crucial to address these growing healthcare needs. The main objective of this study is to develop and validate a population-based prognostic model to predict the probability of unplanned hospitalization in the Basque Country, through comparing the performance of a logistic regression model and three families of machine learning models. METHODS: Using age, sex, diagnoses and drug prescriptions previously transformed by the Johns Hopkins Adjusted Clinical Groups (ACG) System, we predict the probability of unplanned hospitalization in the Basque Country (2.2 million inhabitants) using several techniques. When dealing with non-deterministic algorithms, comparing a single model per technique is not enough to choose the best approach. Thus, we conduct 40 experiments per family of models - Random Forest, Gradient Boosting Decision Trees and Multilayer Perceptrons - and compare them to Logistic Regression. Models' performance are compared both population-wide and for the 20,000 patients with the highest predicted probabilities, as a hypothetical high-risk group to intervene on. RESULTS: The best-performing technique is Multilayer Perceptron, followed by Gradient Boosting Decision Trees, Logistic Regression and Random Forest. Multilayer Perceptrons also have the lowest variability, around an order of magnitude less than Random Forests. Median area under the ROC curve, average precision and positive predictive value range from 0.789 to 0.802, 0.237 to 0.257 and 0.485 to 0.511, respectively. For Brier Score the median values are 0.048 for all techniques. There is some overlap between the algorithms. For instance, Gradient Boosting Decision Trees perform better than Logistic Regression more than 75% of the time, but not always. CONCLUSIONS: All models have good global performance. The only family that is consistently superior to Logistic Regression is Multilayer Perceptron, showing a very reliable performance with the lowest variability.

Multidimensional adaptive P-splines with application to neurons' activity studies.

The receptive field (RF) of a visual neuron is the region of the space that elicits neuronal responses. It can be mapped using different techniques that allow inferring its spatial and temporal properties. Raw RF maps (RFmaps) are usually noisy, making it difficult to obtain and study important features of the RF. A possible solution is to smooth them using P-splines. Yet, raw RFmaps are characterized by sharp transitions in both space and time. Their analysis thus asks for spatiotemporal adaptive P-spline models, where smoothness can be locally adapted to the data. However, the literature lacks proposals for adaptive P-splines in more than two dimensions. Furthermore, the extra flexibility afforded by adaptive P-spline models is obtained at the cost of a high computational burden, especially in a multidimensional setting. To fill these gaps, this work presents a novel anisotropic locally adaptive P-spline model in two (e.g., space) and three (space and time) dimensions. Estimation is based on the recently proposed SOP (Separation of Overlapping Precision matrices) method, which provides the speed we look for. Besides the spatiotemporal analysis of the neuronal activity data that motivated this work, the practical performance of the proposal is evaluated through simulations, and comparisons with alternative methods are reported.

A two-stage approach for the spatio-temporal analysis of high-throughput phenotyping data.

High throughput phenotyping (HTP) platforms and devices are increasingly used for the characterization of growth and developmental processes for large sets of plant genotypes. Such HTP data require challenging statistical analyses in which longitudinal genetic signals need to be estimated against a background of spatio-temporal noise processes. We propose a two-stage approach for the analysis of such longitudinal HTP data. In a first stage, we correct for design features and spatial trends per time point. In a second stage, we focus on the longitudinal modelling of the spatially corrected data, thereby taking advantage of shared longitudinal features between genotypes and plants within genotypes. We propose a flexible hierarchical three-level P-spline growth curve model, with plants/plots nested in genotypes, and genotypes nested in populations. For selection of genotypes in a plant breeding context, we show how to extract new phenotypes, like growth rates, from the estimated genotypic growth curves and their first-order derivatives. We illustrate our approach on HTP data from the PhenoArch greenhouse platform at INRAE Montpellier and the outdoor Field Phenotyping platform at ETH Zürich.

Analysis of the impact of social determinants and primary care morbidity on population health outcomes by combining big data: A research protocol.

Background: In recent years, different tools have been developed to facilitate analysis of social determinants of health (SDH) and apply this to health policy. The possibility of generating predictive models of health outcomes which combine a wide range of socioeconomic indicators with health problems is an approach that is receiving increasing attention. Our objectives are twofold: (1) to predict population health outcomes measured as hospital morbidity, taking primary care (PC) morbidity adjusted for SDH as predictors; and (2) to analyze the geographic variability of the impact of SDH-adjusted PC morbidity on hospital morbidity, by combining data sourced from electronic health records and selected operations of the National Statistics Institute (Instituto Nacional de Estadística/INE). Methods: The following will be conducted: a qualitative study to select socio-health indicators using RAND methodology in accordance with SDH frameworks, based on indicators published by the INE in selected operations; and a quantitative study combining two large databases drawn from different Spain's Autonomous Regions (ARs) to enable hospital morbidity to be ascertained, i.e., PC electronic health records and the minimum basic data set (MBDS) for hospital discharges. These will be linked to socioeconomic indicators, previously selected by geographic unit. The outcome variable will be hospital morbidity, and the independent variables will be age, sex, PC morbidity, geographic unit, and socioeconomic indicators. Analysis: To achieve the first objective, predictive models will be used, with a test-and-training technique, fitting multiple logistic regression models. In the analysis of geographic variability, penalized mixed models will be used, with geographic units considered as random effects and independent predictors as fixed effects. Discussion: This study seeks to show the relationship between SDH and population health, and the geographic differences determined by such determinants. The main limitations are posed by the collection of data for healthcare as opposed to research purposes, and the time lag between collection and publication of data, sampling errors and missing data in registries and surveys. The main strength lies in the project's multidisciplinary nature (family medicine, pediatrics, public health, nursing, psychology, engineering, geography).

Reconstruction of Networks with Direct and Indirect Genetic Effects.

Genetic variance of a phenotypic trait can originate from direct genetic effects, or from indirect effects, i.e., through genetic effects on other traits, affecting the trait of interest. This distinction is often of great importance, for example, when trying to improve crop yield and simultaneously control plant height. As suggested by Sewall Wright, assessing contributions of direct and indirect effects requires knowledge of (1) the presence or absence of direct genetic effects on each trait, and (2) the functional relationships between the traits. Because experimental validation of such relationships is often unfeasible, it is increasingly common to reconstruct them using causal inference methods. However, most current methods require all genetic variance to be explained by a small number of quantitative trait loci (QTL) with fixed effects. Only a few authors have considered the "missing heritability" case, where contributions of many undetectable QTL are modeled with random effects. Usually, these are treated as nuisance terms that need to be eliminated by taking residuals from a multi-trait mixed model (MTM). But fitting such an MTM is challenging, and it is impossible to infer the presence of direct genetic effects. Here, we propose an alternative strategy, where genetic effects are formally included in the graph. This has important advantages: (1) genetic effects can be directly incorporated in causal inference, implemented via our PCgen algorithm, which can analyze many more traits; and (2) we can test the existence of direct genetic effects, and improve the orientation of edges between traits. Finally, we show that reconstruction is much more accurate if individual plant or plot data are used, instead of genotypic means. We have implemented the PCgen-algorithm in the R-package pcgen.

Bootstrap-based procedures for inference in nonparametric receiver-operating characteristic curve regression analysis.

Prior to using a diagnostic test in a routine clinical setting, the rigorous evaluation of its diagnostic accuracy is essential. The receiver-operating characteristic curve is the measure of accuracy most widely used for continuous diagnostic tests. However, the possible impact of extra information about the patient (or even the environment) on diagnostic accuracy also needs to be assessed. In this paper, we focus on an estimator for the covariate-specific receiver-operating characteristic curve based on direct regression modelling and nonparametric smoothing techniques. This approach defines the class of generalised additive models for the receiver-operating characteristic curve. The main aim of the paper is to offer new inferential procedures for testing the effect of covariates on the conditional receiver-operating characteristic curve within the above-mentioned class. Specifically, two different bootstrap-based tests are suggested to check (a) the possible effect of continuous covariates on the receiver-operating characteristic curve and (b) the presence of factor-by-curve interaction terms. The validity of the proposed bootstrap-based procedures is supported by simulations. To facilitate the application of these new procedures in practice, an R-package, known as npROCRegression, is provided and briefly described. Finally, data derived from a computer-aided diagnostic system for the automatic detection of tumour masses in breast cancer is analysed.

Modelling spatial trends in sorghum breeding field trials using a two-dimensional P-spline mixed model.

KEY MESSAGE: A flexible and user-friendly spatial method called SpATS performed comparably to more elaborate and trial-specific spatial models in a series of sorghum breeding trials. Adjustment for spatial trends in plant breeding field trials is essential for efficient evaluation and selection of genotypes. Current mixed model methods of spatial analysis are based on a multi-step modelling process where global and local trends are fitted after trying several candidate spatial models. This paper reports the application of a novel spatial method that accounts for all types of continuous field variation in a single modelling step by fitting a smooth surface. The method uses two-dimensional P-splines with anisotropic smoothing formulated in the mixed model framework, referred to as SpATS model. We applied this methodology to a series of large and partially replicated sorghum breeding trials. The new model was assessed in comparison with the more elaborate standard spatial models that use autoregressive correlation of residuals. The improvements in precision and the predictions of genotypic values produced by the SpATS model were equivalent to those obtained using the best fitting standard spatial models for each trial. One advantage of the approach with SpATS is that all patterns of spatial trend and genetic effects were modelled simultaneously by fitting a single model. Furthermore, we used a flexible model to adequately adjust for field trends. This strategy reduces potential parameter identification problems and simplifies the model selection process. Therefore, the new method should be considered as an efficient and easy-to-use alternative for routine analyses of plant breeding trials.

Comparing Some Production Functions for Inpatient Health Services in Selected Public Hospitals in Spain.

To investigate the adequacy of the widely used Cobb-Douglas and transcendental logarithmic (translog) models of the production functions of hospital inpatient services, the authors fitted these and additive models to data for the four most productive health services of 10 public hospitals in Galicia, Spain (the same four in each). Production, measured as admissions weighted in accordance with their diagnosis-related groups (DRGs), was treated as a function of physician full-time equivalents as surrogate labor factor and number of beds as surrogate capital factor. The results suggest that while the Cobb-Douglas and translog models suffice to represent the production functions of services with low average DRG weight, the greater flexibility of additive models is required for services with higher average DRG weight when only these two inputs are considered.

A new approach to categorising continuous variables in prediction models: Proposal and validation.

When developing prediction models for application in clinical practice, health practitioners usually categorise clinical variables that are continuous in nature. Although categorisation is not regarded as advisable from a statistical point of view, due to loss of information and power, it is a common practice in medical research. Consequently, providing researchers with a useful and valid categorisation method could be a relevant issue when developing prediction models. Without recommending categorisation of continuous predictors, our aim is to propose a valid way to do it whenever it is considered necessary by clinical researchers. This paper focuses on categorising a continuous predictor within a logistic regression model, in such a way that the best discriminative ability is obtained in terms of the highest area under the receiver operating characteristic curve (AUC). The proposed methodology is validated when the optimal cut points' location is known in theory or in practice. In addition, the proposed method is applied to a real data-set of patients with an exacerbation of chronic obstructive pulmonary disease, in the context of the IRYSS-COPD study where a clinical prediction rule for severe evolution was being developed. The clinical variable PCO2 was categorised in a univariable and a multivariable setting.

Nonparametric estimation of time-dependent ROC curves conditional on a continuous covariate.

The receiver-operating characteristic (ROC) curve is the most widely used measure for evaluating the performance of a diagnostic biomarker when predicting a binary disease outcome. The ROC curve displays the true positive rate (or sensitivity) and the false positive rate (or 1-specificity) for different cut-off values used to classify an individual as healthy or diseased. In time-to-event studies, however, the disease status (e.g. death or alive) of an individual is not a fixed characteristic, and it varies along the study. In such cases, when evaluating the performance of the biomarker, several issues should be taken into account: first, the time-dependent nature of the disease status; and second, the presence of incomplete data (e.g. censored data typically present in survival studies). Accordingly, to assess the discrimination power of continuous biomarkers for time-dependent disease outcomes, time-dependent extensions of true positive rate, false positive rate, and ROC curve have been recently proposed. In this work, we present new nonparametric estimators of the cumulative/dynamic time-dependent ROC curve that allow accounting for the possible modifying effect of current or past covariate measures on the discriminatory power of the biomarker. The proposed estimators can accommodate right-censored data, as well as covariate-dependent censoring. The behavior of the estimators proposed in this study will be explored through simulations and illustrated using data from a cohort of patients who suffered from acute coronary syndrome.

Assessment of morbidity in gynaecologic oncology laparoscopy and identification of possible risk factors.

BACKGROUND: The aim of this study was to ascertain the incidence of and the risk factors associated with morbidity in laparoscopy performed on patients with cervical cancer and endometrial cancer. METHODS: This was an observational study of a cohort of 128 women, 89 with endometrial cancer and 39 with cervical cancer from January 2000 to December 2011. We used the Student's t-test or the Mann-Whitney U test for continuous variables, and the Chi-square or Fisher's exact test for categorical variables. RESULTS: Complications were found in 44 patients (34.4%). After a multivariate analysis, among the risk factors associated with the presence of complications as the only type of surgery was found to be statistically significant (p = 0.043), more frequent in the most complex procedures such as Wertheim operation, trachelectomy, and para-aortic lymphadenectomy. Type of surgery (p = 0.003) and tumour type (p = 0.003) were risk factors associated with conversion to laparotomy. It was more frequent among the most complex procedures and cervical cancer cases. Regarding the need for transfusion, significant differences were observed in terms of surgery duration (p < 0.001), more frequent in longer surgery. CONCLUSION: Morbidity in laparoscopic surgical oncology is related to the surgery complexity, where the basal characteristics of the patient are not a factor of influence in the development of complications.

ME20-S as a Potential Biomarker for the Evaluation of Uveal Melanoma.

PURPOSE: We previously identified the presence of the melanocyte-specific secreted (ME20-S) glycoprotein in secretomes of uveal melanoma (UM) cultures. The aim of this study was to test for the presence and levels of ME20-S in the serum of patients with choroidal nevi and UM and correlate these levels with individual clinical data. METHODS: Serum ME20-S levels were determined by ELISA in 111 patients distributed into four categories (53 choroidal nevi, 30 untreated UM, 11 10-year disease-free [DF] UM, 17 hepatic metastatic UM) and 32 age- and sex-matched controls. ME20-S levels were correlated with individual clinical data. RESULTS: The UM and the metastatic groups showed significantly higher levels of serum ME20-S than the other groups (P < 0.001). ME20-S levels in the DF patients did not differ from those in the control group. In addition, log-transformed serum ME20-S levels showed a positive correlation with the thickness of the lesion mass in UM patients (regression coefficient 0.0689, 95% confidence interval 0.0689-0.1123, R2 = 27.1%). CONCLUSIONS: Elevated ME20-S serum levels are associated with tumor size and advanced stages of UM while low levels are characteristic of DF patients. ME20-S might be a promising serum marker for UM and useful for monitoring metastatic disease.

Prevalence and genetic diversity of Salmonella spp. in a river in a tropical environment in Mexico.

The capability of Salmonella to survive outside a host is especially relevant in tropical regions, where the environmental conditions could be more suitable for its long-term persistence. This study investigated the prevalence and genetic diversity of salmonellae within rivers of the Culiacan Valley in the northwestern region of Mexico. From July 2008 to June 2009, a total of 138 water samples were evaluated for the presence of Salmonella spp.; additionally, its association with environmental parameters was determined using Generalized Additive Models (GAMs). Salmonella spp. were isolated from 111 (80.4%) samples without any statistical influence on the environmental parameters investigated, according to the GAM analysis. Twenty-four serotypes were identified; the most frequently isolated serotypes were Salmonella Oranienburg (25%), Salmonella Saintpaul (9%) and Salmonella Minnesota (6%). Diverse genetic variants of Salmonella Oranienburg were found distributed across the valley with no distinctive geographical or temporal patterns. The high persistence of Salmonella spp. and the lack of differentiation of types found along the river basins suggest the existence of non-point source contamination. Furthermore, the discrepancy between the prevailing serotypes in human infections and those identified in this study denotes a limited influence of these aquatic environments in bacterial dissemination and disease transmission.

Relationship between macular bending and foveoschisis in myopic patients.

PURPOSE: To investigate factors associated with myopic foveoschisis and macular bending and to determine how the presence of macular bending affects the development of myopic foveoschisis. METHODS: In a prospective study of 194 eyes of 105 patients with high myopia, we performed complete ophthalmic examinations, optical coherence tomography (OCT), and A-scan ultrasounds. Patients were divided into three groups according to the OCT results. Group 1 consisted of 25 eyes (17 patients) with myopic foveoschisis; group 2 consisted of 36 eyes (20 patients) with macular bending; and group 3 consisted of 135 eyes (68 patients) without macular bending, foveoschisis, or other diseases. Macular bending was defined as a smooth macular elevation observed upon OCT in patients with pathologic myopia. Age, sex, spherical equivalence, axial length (AXL), and OCT findings were obtained and compared to identify factors that are related to myopic foveoschisis and macular bending. Moreover, using the whole data set, we evaluated and correlated myopic foveoschisis with the presence or absence of macular bending to determine whether this bulge in the macular area influences the development of myopic foveoschisis. RESULTS: In group 1, all eyes presented posterior staphyloma and two factors were independently associated with a higher risk of having myopic foveoschisis: internal limiting membrane detachment (p < 0.001) and retinal arteriolar traction (p < 0.001). In group 2, only retinal arteriolar traction (p < 0.009) was independently associated with macular bending. Furthermore, macular bending was significantly correlated as a protective factor against myopic foveoschisis (adjusted odds ratio, 0.116; 95% confidence interval, 0.019 to 0.701; p < 0.019); the AXL of patients with the same spherical equivalence and macular bending was significantly shorter than that of patients without macular bending (p = 0.005). CONCLUSIONS: Intraocular and extraocular wall factors were associated with myopic traction maculopathy, which plays an important role in its pathogenesis. Moreover, macular bending might be a key factor in preventing myopic foveoschisis by decreasing AXL.

Insulin resistance (HOMA-IR) cut-off values and the metabolic syndrome in a general adult population: effect of gender and age: EPIRCE cross-sectional study.

BACKGROUND: Insulin resistance has been associated with metabolic and hemodynamic alterations and higher cardio metabolic risk. There is great variability in the threshold homeostasis model assessment of insulin resistance (HOMA-IR) levels to define insulin resistance. The purpose of this study was to describe the influence of age and gender in the estimation of HOMA-IR optimal cut-off values to identify subjects with higher cardio metabolic risk in a general adult population. METHODS: It included 2459 adults (range 20-92 years, 58.4% women) in a random Spanish population sample. As an accurate indicator of cardio metabolic risk, Metabolic Syndrome (MetS), both by International Diabetes Federation criteria and by Adult Treatment Panel III criteria, were used. The effect of age was analyzed in individuals with and without diabetes mellitus separately. ROC regression methodology was used to evaluate the effect of age on HOMA-IR performance in classifying cardio metabolic risk. RESULTS: In Spanish population the threshold value of HOMA-IR drops from 3.46 using 90th percentile criteria to 2.05 taking into account of MetS components. In non-diabetic women, but no in men, we found a significant non-linear effect of age on the accuracy of HOMA-IR. In non-diabetic men, the cut-off values were 1.85. All values are between 70th-75th percentiles of HOMA-IR levels in adult Spanish population. CONCLUSIONS: The consideration of the cardio metabolic risk to establish the cut-off points of HOMA-IR, to define insulin resistance instead of using a percentile of the population distribution, would increase its clinical utility in identifying those patients in whom the presence of multiple metabolic risk factors imparts an increased metabolic and cardiovascular risk. The threshold levels must be modified by age in non-diabetic women.

Serum DJ-1/PARK 7 is a potential biomarker of choroidal nevi transformation.

PURPOSE: There is substantial evidence that intraocular melanomas arise from benign nevi in the uveal tract. Previous studies performed by the authors revealed that uveal melanoma cells secrete the oncoprotein DJ-1/PARK7 into the extracellular environment and circulation. The aim of this study was to determine whether circulating DJ-1 serum levels correlate with known clinical risk factors of nevi growth. METHODS: Standardized ultrasonography, optical coherence tomography, and eye fundus examinations were used to evaluate the clinical risk factors of nevi growth. These clinical risk factors (including nevi size, distance of margins to the optic disc, detection of acoustic hollowness, presence of ocular symptoms, orange pigment, subretinal fluid, and absence of drusen) were examined in 53 consecutive patients from January 2009 to February 2011. Serum levels of DJ-1/PARK7 in these patients and in healthy age- and sex-matched controls (n = 32) were analyzed using ELISA. RESULTS: Within the choroidal nevi group, DJ-1 serum levels were higher in those with symptoms (P < 0.033), with a nevus thickness greater than 1.5 mm (P < 0.001), a large basal diameter greater than 8 mm (P < 0.001), and the presence of acoustic hollowness (P < 0.001), compared to those patients without these risk factors. Similar significant differences were found when these at risk nevi subgroups were compared to healthy persons. CONCLUSIONS: Elevated serum levels of DJ-1 are associated with choroidal nevi transformation risk factors. Therefore, DJ-1 appears to be a promising factor for predicting the growth of choroidal nevi and may be a potential biomarker of malignancy.

Insulin resistance index (HOMA-IR) levels in a general adult population: curves percentile by gender and age. The EPIRCE study.

AIMS: To describe the distribution of HOMA-IR levels in a general nondiabetic population and its relationships with metabolic and lifestyles characteristics. METHODS: Cross-sectional study. Data from 2246 nondiabetic adults in a random Spanish population sample, stratified by age and gender, were analyzed. Assessments included a structured interview, physical examination, and blood sampling. Generalized additive models (GAMs) were used to assess the effect of lifestyle habits and clinical and demographic measurements on HOMA-IR. Multivariate GAMs and quantile regression analyses of HOMA-IR were carried out separately in men and women. RESULTS: This study shows refined estimations of HOMA-IR levels by age, body mass index, and waist circumference in men and women. HOMA-IR levels were higher in men (2.06) than women (1.95) (P=0.047). In women, but not men, HOMA-IR and age showed a significant nonlinear association (P=0.006), with increased levels above fifty years of age. We estimated HOMA-IR curves percentile in men and women. CONCLUSIONS: Age- and gender-adjusted HOMA-IR levels are reported in a representative Spanish adult non-diabetic population. There are gender-specific differences, with increased levels in women over fifty years of age that may be related with changes in body fat distribution after menopause.

Admission and fasting plasma glucose for estimating risk of death of diabetic and nondiabetic patients with acute coronary syndrome: nonlinearity of hazard ratios and time-dependent comparison.

BACKGROUND: In patients with acute coronary syndrome (ACS), increased plasma glucose levels are associated with worse outcome. Our aim is to ascertain the values of admission and fasting glucose for prediction of death among patients with ACS; and to compare their predictive capacities. METHODS: The relationships of mortality to plasma glucose levels among 811 consecutive patients hospitalized with ACS were estimated using spline Cox models. Blood samples were obtained upon admission and after overnight fast. The predictive capacities of fasting and admission glucose were compared using time-dependent receiver operating characteristic curves. RESULTS: Fasting and admission glucose levels were higher among the 151 patients who died (18.6%) than among survivors (P < .001). Among the 558 patients with no history of diabetes (68.8%) there was a J-shaped dependence of the all-time mortality hazard ratio on fasting glucose that persisted when adjusted for covariates: hazard was lowest at 110 mg/dL (6.1 mmol/L), and significantly greater at levels <90 mg/dL (5.0 mmol/L) or >117 mg/dL (6.5 mmol/L). Likewise among non-diabetic patients, the predictive capacities of admission and fasting glucose were similar for forecast times of up to about 1 year, but for later times the area under the receiver operating characteristic curve was larger for fasting glucose than admission glucose (P < .05). Neither admission nor fasting glucose levels discriminated among diabetic patients in regard to risk of death. CONCLUSIONS: Both admission and fasting glucose may be used for triage of nondiabetic ACS patients; fasting glucose may additionally be useful for long-term management, for which the relationship with the all-time mortality hazard ratio is J-shaped.