A hospital wide predictive model for unplanned readmission using hierarchical ICD data.

BACKGROUND AND OBJECTIVE: Hospitals already acquire a large amount of data, mainly for administrative, billing and registration purposes. Tapping on these already available data for additional purposes, aiming at improving care, without significant incremental effort and cost. This potential of secondary patient data is explored through modeling administrative and billing data, as well as the hierarchical structure of pathology codes of the International Classification of Diseases (ICD) in the prediction of unplanned readmissions, as a clinically relevant outcome parameter that can be impacted on in a quality improvement program. METHODS: In this single-center, hospital-wide observational cohort study, we included all adult patients discharged in 2016 after applying an exclusion protocol (n = 29,702). In addition to administrative variables, such as age and length of stay, structured pathology data were taken into account in predictive models. As a first research question, we compared logistic regression against penalized logistic regression, gradient boosting and Random Forests to predict unplanned readmission. As a second research goal, we investigated the level of hierarchy within the pathology data needed to achieve the best accuracy. Finally, we investigated which prediction variables play a prominent role in predicting hospital readmission. The performance of all models was evaluated using the Area Under the ROC Curve (AUC) measure. RESULTS: All models have the best predictive results using Random Forests. An added value of 7% is observed compared to a baseline method such as logistic regression. The best model, based on Random Forests, achieved an AUC of 0.77, using the diagnosis category and procedure code as lowest level of the hierarchical pathology data. CONCLUSIONS: The most accurate model to predict hospital wide unplanned readmission is based on Random Forests and includes the ICD hierarchy, especially diagnosis category. Such an approach lowers the number of predictor variables and yields a higher interpretability than a model based on a detailed diagnosis. The performance of the model proved high enough to be used as a decision support tool.

Novel approaches to assess the quality of fertility data stored in dairy herd management software.

Scientific journals and popular press magazines are littered with articles in which the authors use data from dairy herd management software. Almost none of such papers include data cleaning and data quality assessment in their study design despite this being a very critical step during data mining. This paper presents 2 novel data cleaning methods that permit identification of animals with good and bad data quality. The first method is a deterministic or rule-based data cleaning method. Reproduction and mutation or life-changing events such as birth and death were converted to a symbolic (alphabetical letter) representation and split into triplets (3-letter code). The triplets were manually labeled as physiologically correct, suspicious, or impossible. The deterministic data cleaning method was applied to assess the quality of data stored in dairy herd management from 26 farms enrolled in the herd health management program from the Faculty of Veterinary Medicine Ghent University, Belgium. In total, 150,443 triplets were created, 65.4% were labeled as correct, 17.4% as suspicious, and 17.2% as impossible. The second method, a probabilistic method, uses a machine learning algorithm (random forests) to predict the correctness of fertility and mutation events in an early stage of data cleaning. The prediction accuracy of the random forests algorithm was compared with a classical linear statistical method (penalized logistic regression), outperforming the latter substantially, with a superior receiver operating characteristic curve and a higher accuracy (89 vs. 72%). From those results, we conclude that the triplet method can be used to assess the quality of reproduction data stored in dairy herd management software and that a machine learning technique such as random forests is capable of predicting the correctness of fertility data.

RECIPE COMPLETION USING MACHINE LEARNING TECHNIQUES.

Completing a recipe is a non-trivial task, as the success of ingredient combinations depends on a multitude of factors such as taste, smell, texture, etc. The aim of our work is to build a model that adds one or more ingredients to a given number of ingredients. The idea is based on leftover ingredients in a fridge. A person could list the available ingredients in his or her fridge and the model would suggest some additional ingredients to create a full recipe.

The logistic curve as a tool to describe the daily ruminal pH pattern and its link with milk fatty acids.

Daily ruminal pH variation can be summarized by a cumulative logistic curve based on the amount of time below multiple pH points and characterized by 2 parameters (ß(0) and ß(1)). Moreover, rumen pH variation affects the rumen microbiome as well as the biohydrogenation pathways resulting in a modified secretion of milk fatty acids (FA). The aims of this study were to assess the shifts in milk FA due to rumen pH changes and to estimate the relationship between milk FA and the 2 parameters of the logistic curve. The data consisted of milk samples of 2 experiments. In experiment 1, 3 cows were subjected to 5 treatments in which the type and amount of concentrate were changed during 33 d: (1) control diet 1, (2) stepwise replacement of a standard concentrate (CONC) by a CONC rich in rapidly fermentable carbohydrates, (3) increase in the total amount of CONC, (4) treatment with a buffer solution, and (5) control diet 2. A 3×3 Latin square design with 3 cows was used in the second experiment. During the first 14 d of each period, the cows received a control diet with a standard CONC, whereas in the last 7 d the standard CONC was replaced step-by-step by a CONC rich in rapidly fermentable carbohydrates and the amount of CONC was increased. During each period, a different buffer treatment was added to the diet. Milk FA and pH reacted similarly in both experiments: decreasing proportions of iso FA and increasing proportions of odd-chain FA were observed. However, an abrupt change to a 76% CONC diet as for one cow of experiment 1 led to almost a 10-fold increase in C18:1 trans-10 (0.79 vs. 6.75 g/100g of FA). In experiment 2, the stepwise approach of adding CONC and the continuous supplementation of buffer led to minimal increases in C18:1 trans-10 and decreases in rumen pH compared with the diet with standard CONC only. Fatty acid proportions were influenced by the level of rumen pH (ß(1)) or the rumen pH variation (ß(0)), or both. High proportions of C18:1 trans-10 (above 4 g/100g of FA) occurred with low and largely fluctuating pH (low ß(1), low ß(0)), whereas situations with low, stable pH (low ß(1), great ß(0)) did not induce a shift toward the secondary biohydrogenation pathway. C18:1 trans-11 and C18:2 cis-9, trans-11 were only influenced by the pH variation and not by the average pH, whereas iso C14:0 and iso C16:0 FA were only dependent on the average pH and not influenced by diurnal pH variation. Overall, milk FA changes were related to pH changes; however, this relationship is not straightforward and needs further research.

Toward a Reliable Evaluation of Forecasting Systems for Plant Diseases: A Case Study Using Fusarium Head Blight of Wheat.

Despite great efforts to forecast plant diseases, many of the existing systems often fall short in providing farmers with accurate predictions. One of the main problems arises from the existence of year and location effects, so that more advanced procedures are required for evaluating existing systems in an unbiased manner. This paper illustrates the case of Fusarium head blight of winter wheat in Belgium. We present a new cross-validation strategy that enables the evaluation of the predictive performance of a forecasting system for years and locations that are different from the years and locations on which the forecast was developed. Four different cross-validation strategies and five regression techniques are used. The results demonstrated that traditional evaluation strategies are too optimistic in their predictions, whereas the cross-year cross-location validation strategy yielded more realistic outcomes. Using this procedure, the mean squared error increased and the coefficient of determination decreased in predicting disease severity and deoxynivalenol content, suggesting that existing evaluation strategies may generate a substantial optimistic bias. The strongest discrepancies between the cross-validation strategies were observed for multiple linear regression models.

Polymorphisms in the ficolin 1 gene (FCN1) are associated with susceptibility to the development of rheumatoid arthritis.

OBJECTIVES: We investigated the possible association of rheumatoid arthritis (RA) with single nucleotide polymorphisms (SNP) within the ficolin (FCN) genes. Two SNPs in the FCN1 gene, four SNPs in the FCN2 gene and one SNP in the FCN3 gene were studied. METHODS: The SNPs within the FCN genes were detected by an experimental INNO-LiPA methodology (Innogenetics, Belgium) in a population consisting of 338 RA patients and 595 controls. The significant SNPs were further evaluated in two subpopulations and related to carriage of the human leukocyte antigen-shared epitope (HLA-SE), rheumatoid factor (RF) and the presence of anti-citrullinated protein/peptide antibodies (ACPA). RESULTS: Two SNPs in the FCN1 gene were significantly associated with RA: the A allele rs2989727 was significantly increased in RA patients (67%) compared with controls (60%) (P = 0.002). Also, the frequency of the G allele of rs1071583 was increased in RA patients (68%) compared with controls (61%) (P = 0.003). Analysis of agreement between SNPs suggested strong linkage between rs2989727 and rs1071583. Carriage of a FCN1 SNP was independent of carriage of the HLA-SE, RF status and ACPA positivity. CONCLUSIONS: We describe two linked SNPs in the FCN1 gene that are associated with the development of RA.