Validation and development of a clinical prediction rule in clinically suspected community-acquired pneumonia

To develop a mathematical model to predict the probability of having community-acquired pneumonia and to evaluate an already developed prediction rule that has not been validated in a clinical scenario.

Stable feature selection for clinical prediction: Exploiting ICD tree structure using Tree-Lasso

Modern healthcare is getting reshaped by growing Electronic Medical Records (EMR). Recently, these records have been shown of great value towards building clinical prediction models. In EMR data, patients' diseases and hospital interventions are captured through a set of diagnoses and procedures codes. These codes are usually represented in a tree form (e.g. ICD-10 tree) and the codes within a tree branch may be highly correlated. These codes can be used as features to build a prediction model and an appropriate feature selection can inform a clinician about important risk factors for a disease. Traditional feature selection methods (e.g. Information Gain, T-test, etc.) consider each variable independently and usually end up having a long feature list. Recently, Lasso and related l1-penalty based feature selection methods have become popular due to their joint feature selection property. However, Lasso is known to have problems of selecting one feature of many correlated features randomly. This hinders the clinicians to arrive at a stable feature set, which is crucial for clinical decision making process. In this paper, we solve this problem by using a recently proposed Tree-Lasso model. Since, the stability behavior of Tree-Lasso is not well understood, we study the stability behavior of Tree-Lasso and compare it with other feature selection methods. Using a synthetic and two real-world datasets (Cancer and Acute Myocardial Infarction), we show that Tree-Lasso based feature selection is significantly more stable than Lasso and comparable to other methods e.g. Information Gain, ReliefF and T-test. We further show that, using different types of classifiers such as logistic regression, naive Bayes, support vector machines, decision trees and Random Forest, the classification performance of Tree-Lasso is comparable to Lasso and better than other methods. Our result has implications in identifying stable risk factors for many healthcare problems and therefore can potentially assist clinical decision making for accurate medical prognosis.

Variable definitions: implications for the prediction of pulmonary complications after adult cardiac surgery

Aim: The aim of this paper was to review the implications that variable definitions have for the prediction of post-operative pulmonary complications after cardiac surgery.

Method: A review of the literature from 1980 to 2002. Selected studies demonstrated an original attempt to examine multivariate associations between pre, intra or post-operative antecedents and pulmonary outcomes in patients undergoing coronary artery bypass grafting (CABG). Reports that described the validation of established clinical prediction rules, testing interventions or research conducted in non-human cohorts were excluded from this review.

Results: Consistently, variable factor and outcome definitions are combined for the development of multivariate prediction models that subsequently have limited clinical value. Despite being prevalent there are very few attempts to examine post-operative pulmonary complications (PPC) as endpoints in isolation. The trajectory of pulmonary dysfunction that precedes complications in the post-operative context is not clear. As such there is little knowledge of post-operative antecedents to PPC that are invariably excluded from model development.

Conclusion: Multivariate clinical prediction rules that incorporate antecedent patient and process factors from the continuum of cardiovascular care for specific pulmonary outcomes are recommended. Models such as these would be useful for practice, policy and quality improvement.

Stabilizing l1-norm prediction models by supervised feature grouping

Emerging Electronic Medical Records (EMRs) have reformed the modern healthcare. These records have great potential to be used for building clinical prediction models. However, a problem in using them is their high dimensionality. Since a lot of information may not be relevant for prediction, the underlying complexity of the prediction models may not be high. A popular way to deal with this problem is to employ feature selection. Lasso and l1-norm based feature selection methods have shown promising results. But, in presence of correlated features, these methods select features that change considerably with small changes in data. This prevents clinicians to obtain a stable feature set, which is crucial for clinical decision making. Grouping correlated variables together can improve the stability of feature selection, however, such grouping is usually not known and needs to be estimated for optimal performance. Addressing this problem, we propose a new model that can simultaneously learn the grouping of correlated features and perform stable feature selection. We formulate the model as a constrained optimization problem and provide an efficient solution with guaranteed convergence. Our experiments with both synthetic and real-world datasets show that the proposed model is significantly more stable than Lasso and many existing state-of-the-art shrinkage and classification methods. We further show that in terms of prediction performance, the proposed method consistently outperforms Lasso and other baselines. Our model can be used for selecting stable risk factors for a variety of healthcare problems, so it can assist clinicians toward accurate decision making.

Stabilizing sparse cox model using statistic and semantic structures in electronic medical records

Stability in clinical prediction models is crucial for transferability between studies, yet has received little attention. The problem is paramount in high dimensional data, which invites sparse models with feature selection capability. We introduce an effective method to stabilize sparse Cox model of time-to-events using statistical and semantic structures inherent in Electronic Medical Records (EMR). Model estimation is stabilized using three feature graphs built from (i) Jaccard similarity among features (ii) aggregation of Jaccard similarity graph and a recently introduced semantic EMR graph (iii) Jaccard similarity among features transferred from a related cohort. Our experiments are conducted on two real world hospital datasets: a heart failure cohort and a diabetes cohort. On two stability measures – the Consistency index and signal-to-noise ratio (SNR) – the use of our proposed methods significantly increased feature stability when compared with the baselines.

Guidelines for developing and reporting machine learning predictive models in biomedical research : a multidisciplinary view

BACKGROUND: As more and more researchers are turning to big data for new opportunities of biomedical discoveries, machine learning models, as the backbone of big data analysis, are mentioned more often in biomedical journals. However, owing to the inherent complexity of machine learning methods, they are prone to misuse. Because of the flexibility in specifying machine learning models, the results are often insufficiently reported in research articles, hindering reliable assessment of model validity and consistent interpretation of model outputs.

OBJECTIVE: To attain a set of guidelines on the use of machine learning predictive models within clinical settings to make sure the models are correctly applied and sufficiently reported so that true discoveries can be distinguished from random coincidence.

METHODS: A multidisciplinary panel of machine learning experts, clinicians, and traditional statisticians were interviewed, using an iterative process in accordance with the Delphi method.

RESULTS: The process produced a set of guidelines that consists of (1) a list of reporting items to be included in a research article and (2) a set of practical sequential steps for developing predictive models.

CONCLUSIONS: A set of guidelines was generated to enable correct application of machine learning models and consistent reporting of model specifications and results in biomedical research. We believe that such guidelines will accelerate the adoption of big data analysis, particularly with machine learning methods, in the biomedical research community.

Fracture prediction from repeat BMD measurements in clinical practice

Summary: We investigated whether repeat BMD measurements in clinical populations are useful for fracture risk assessment. We report that repeat BMD measurements are a robust predictor of fracture in clinical populations; this is not affected by preceding BMD change or recent osteoporosis therapy. Introduction: In clinical practice, many patients selectively undergo repeat bone mineral density (BMD) measurements. We investigated whether repeat BMD measurements in clinical populations are useful for fracture risk assessment and whether this is affected by preceding change in BMD or recent osteoporosis therapy. Methods: We identified women and men aged ≥50 years who had a BMD measurement during 1990–2009 from a large clinical BMD database for Manitoba, Canada (n = 50,215). Patient subgroups aged ≥50 years at baseline with repeat BMD measures were identified. Data were linked to an administrative data repository, from which osteoporosis therapy, fracture outcomes, and covariates were extracted. Using Cox proportional hazards models, we assessed covariate-adjusted risk for major osteoporotic fracture (MOF) and hip fracture according to BMD (total hip, lumbar spine, femoral neck) at different time points. Results: Prevalence of osteoporosis therapy increased from 18 % at baseline to 55 % by the fourth measurement. Total hip BMD was predictive of MOF at each time point. In the patient subgroup with two repeat BMD measurements (n = 13,481), MOF prediction with the first and second measurements was similar: adjusted-hazard ratio (HR) per SD 1.45 (95 % CI 1.34–1.56) vs. 1.64 (95 % CI 1.48–1.81), respectively. No differences were seen when the second measurement results were stratified by preceding change in BMD or osteoporosis therapy (both p-interactions >0.2). Similar results were seen for hip fracture prediction and when spine and femoral neck BMD were analyzed. Conclusion: Repeat BMD measurements are a robust predictor of fracture in clinical populations; this is not affected by preceding BMD change or recent osteoporosis therapy.

The clinical application of three screening tools for recognizing post-partum depression

Postnatal depression is a major health issue for childbearing women world-wide, as it is not always identified early. This study aimed to evaluate the clinical application of three screening instruments for the early recognition of post-partum depression, the Postpartum Depression Prediction Inventory, the Postpartum Depression Screening Scale and the Edinburgh Postnatal Depression Scale, and to examine nurse interventions following use of the instruments. Data were collected at two points, at 28 weeks prenatal (107 women) and eight weeks postnatal (84 women). Results showed that 17% of the women scored significant symptoms of post-partum depression and 10–15% had a positive screen for major postnatal depression. There was a statistically significant correlation between the total score on the Postpartum Depression Screening Scale and the Edinburgh Postnatal Depression Scale. Of those eight women identified as being at risk, seven had received anticipatory guidance and five had received counselling by the nurses. The Postpartum Depression Prediction Inventory enabled nurses to identify women at risk of post-partum depression and offer interventions.

Copper-transporting P-Type ATPase, ATP7A, confers multidrug resistance and its expression is related to resistance to SN-38 in clinical colon cancer

We and others have shown that the copper transporters ATP7A and ATP7B play a role in cellular resistance to cisdiaminedichloroplatinum (II) (CDDP).  In this study, we found that ATP7A transfection of Chinese hamster ovary  cells (CHOK1) and fibroblasts isolated from Menkes disease patients  enhanced resistance not only to CDDP but also to various anticancer drugs, such as vincristine, paclitaxel, 7-ethyl-10- hydroxy-camptothecin (SN-38),  etoposide, doxorubicin, mitoxantron, and 7-ethyl-10-[4-(1-piperidino)-1-piperidino] carbonyloxycamptothecin (CPT-11). ATP7A preferentially localized
doxorubicin fluorescence to the Golgi apparatus in contrast to the more intense nuclear staining of doxorubicin in the parental cells. Brefeldin A   partially and monensin completely altered the distribution of doxorubicin to the nuclei in the ATP7A-expressing cells. ATP7A expression also enhanced the efflux rates of doxorubicin and SN-38 from cells and increased the uptake of SN-38 in membrane vesicles. These findings strongly suggested that   ATP7A confers multidrug resistance to the cells by compartmentalizing drugs in the Golgi apparatus and by enhancing efflux of these drugs, and the trans-Golgi network has an important role of ATP7A-related drug resistance. ATP7A was expressed in 8 of 34 (23.5%) clinical colon cancer specimens but not in the adjacent normal epithelium. Using the histoculture drug response assay that is useful for the prediction of drug sensitivity of clinical cancers, ATP7A-expressing colon cancer cells were significantly more  resistant to SN-38 than ATP7Anegative cells. Thus, ATP7A confers  resistance to various anticancer agents on cancer cells and might be a good index of drug resistance in clinical colon cancers.

Kernel-based naive bayes classifier for breast cancer prediction

The classification of breast cancer patients is of great importance in cancer diagnosis. Most classical cancer classification methods are clinical-based and have limited diagnostic ability. The recent advances in machine learning technique has made a great impact in cancer diagnosis. In this research, we develop a new algorithm: Kernel-Based Naive Bayes (KBNB) to classify breast cancer tumor based on memography data. The performance of the proposed algorithm is compared with that of classical navie bayes algorithm and kernel-based decision tree algorithm C4.5. The proposed algorithm is found to outperform in the both cases. We recommend the proposed algorithm could be used as a tool to classify the breast patient for early cancer diagnosis.

Cardiovascular risk prediction model development and assessment

The aim of this article is to review the development and assessment of cardiovascular risk prediction models and to discuss the predictive value of a risk factor as well as to introduce new assessment methods to evaluate a risk prediction model. Many cardiovascular risk prediction models have been developed during the past three decades. However, there has not been consistent agreement regarding how to appropriately assess a risk prediction model, especially when new markers are added to an existing model. The area under the receiver operating characteristic (ROC) curve has traditionally been used to assess the discriminatory ability of a risk prediction model. However, recent studies suggest that this method has its limitations and cannot be the sole approach to evaluate the usefulness of a new marker. New assessment methods are being developed to appropriately assess a risk prediction model and they will be gradually used in clinical and epidemiological studies.

Overview of risk prediction models in cardiovascular disease research

Many risk prediction models have been developed for cardiovascular diseases in different countries during the past three decades. However, there has not been consistent agreement regarding how to appropriately assess a risk prediction model, especially when new markers are added to an established risk prediction model. Researchers often use the area under the receiver operating characteristic curve (ROC) to assess the discriminatory ability of a risk prediction model. However, recent studies suggest that this method has serious limitations and cannot be the sole approach to evaluate the usefulness of a new marker in clinical and epidemiological studies. To overcome the shortcomings of this traditional method, new assessment methods have been proposed. The aim of this article is to overview various risk prediction models for cardiovascular diseases, to describe the receiver operating characteristic curve method and discuss some new assessment methods proposed recently. Some of the methods were illustrated with figures from a cardiovascular disease study in Australia.

Clinical, ethical and legal considerations in the treatment of newborns with non-ketotic hyperglycinaemia

Non-ketotic hyperglycinaemia (NKH) is a devastating neurometabolic disorder leading, in its classical form, to early death or severe disability and poor quality of life in survivors. Affected neonates may need ventilatory support during a short period of respiratory depression. The transient dependence on ventilation dictates urgency in decision-making regarding withdrawal of therapy. The occurrence of patients with apparent transient forms of the disease, albeit rare, adds uncertainty to the prediction of clinical outcome and dictates that the current practice of withholding or withdrawing therapy in these neonates be reviewed. Both bioethics and law take the view that treatment decisions should be based on the best interests of the patient. The medical-ethics approach is based on the principles of non-maleficence, beneficence, autonomy and justice. The law relating to withholding or withdrawing life-sustaining treatment is complex and varies between jurisdictions. Physicians treating newborns with NKH need to provide families with accurate and complete information regarding the disease and the relative probability of possible outcomes of the neonatal presentation and to explore the extent to which family members are willing to take part in the decision making process. Cultural and religious attitudes, which may potentially clash with bioethical and juridical principles, need to be considered.