Use of Machine Learning Classifiers and Sensor Data to Detect Neurological Deficit in Stroke Patients

Artificial Intelligence and Heart-Brain Connections: A Narrative Review on Algorithms Utilization in Clinical Practice

Cardiovascular and neurological diseases are a major cause of mortality and morbidity worldwide. Such diseases require careful monitoring to effectively manage their progression. Artificial intelligence (AI) offers valuable tools for this purpose through its ability to analyse data and identify predictive patterns. This review evaluated the application of AI in cardiac and neurological diseases for their clinical impact on the general population. We reviewed studies on the application of AI in the neurological and cardiological fields. Our search was performed on the PubMed, Web of Science, Embase and Cochrane library databases. Of the initial 5862 studies, 23 studies met the inclusion criteria. The studies showed that the most commonly used algorithms in these clinical fields are Random Forest and Artificial Neural Network, followed by logistic regression and Support-Vector Machines. In addition, an ECG-AI algorithm based on convolutional neural networks has been developed and has been widely used in several studies for the detection of atrial fibrillation with good accuracy. AI has great potential to support physicians in interpretation, diagnosis, risk assessment and disease management.

Machine learning to detect, stage and classify diseases and their symptoms based on inertial sensor data: a mapping review

This article presents a systematic review aimed at mapping the literature published in the last decade on the use of machine learning (ML) for clinical decision-making through wearable inertial sensors. The review aims to analyze the trends, perspectives, strengths, and limitations of current literature in integrating ML and inertial measurements for clinical applications. The review process involved defining four research questions and applying four relevance assessment indicators to filter the search results, providing insights into the pathologies studied, technologies and setups used, data processing schemes, ML techniques applied, and their clinical impact. When combined with ML techniques, inertial measurement units (IMUs) have primarily been utilized to detect and classify diseases and their associated motor symptoms. They have also been used to monitor changes in movement patterns associated with the presence, severity, and progression of pathology across a diverse range of clinical conditions. ML models trained with IMU data have shown potential in improving patient care by objectively classifying and predicting motor symptoms, often with a minimally encumbering setup. The findings contribute to understanding the current state of ML integration with wearable inertial sensors in clinical practice and identify future research directions. Despite the widespread adoption of these technologies and techniques in clinical applications, there is still a need to translate them into routine clinical practice. This underscores the importance of fostering a closer collaboration between technological experts and professionals in the medical field.

Recombination of the right cerebral cortex in patients with left side USN after stroke: fNIRS evidence from resting state

Objective

Unilateral spatial neglect (USN) is an impaired contralesional stimulus detection, response, or action, causing functional disability. After a stroke, the right hemisphere experiences USN more noticeably, severely, and persistently than the left. However, few studies using fNIRS have been reported in cases of USN. This study aimed to confirm weaker RSFC in USN and investigate the potential inherent features in hemodynamic fluctuations that may be associated with USN. Furthermore, these features were combined into a mathematical model for more accurate classification.

Methods

A total of 33 stroke patients with right-sided brain damage were chosen, of whom 12 had non-USN after stroke, and 21 had USN after stroke (the USN group). Graph theory was used to evaluate the hemodynamic signals of the brain's right cerebral cortex during rest. Furthermore, a support vector machine model was built to categorize the subjects into two groups based on the chosen network properties.

Results

First, mean functional connectivity was lower in the USN group (0.745 ± 0.239) than in the non-USN group (0.843 ± 0.254) (t = -4.300, p < 0.001). Second, compared with the non-USN group, USN patients had a larger clustering coefficient (C) (t = 3.145, p < 0.001), local efficiency (LE) (t = 3.189, p < 0.001), and smaller global efficiency (GE) (t = 3.047, p < 0.001). Notably, there were differences in characteristic path length (L) and small worldness (σ) values between the two groups at certain thresholds, mainly as higher L (t = 3.074, p < 0.001) and lower small worldness (σ) values (t = 2.998, p < 0.001) in USN patients compared with non-USN patients. Finally, the classification accuracy of the SVM model based on AUC aC (t = -2.259, p = 0.031) and AUC aLE (t = -2.063, p = 0.048) was 85%, the sensitivity was 75%, and the specificity was 89%.

Conclusion

The functional network architecture of the right cerebral cortex exhibits significant topological alterations in individuals with USN following stroke, and the sensitivity index based on the small-world property AUC may be utilized to identify these patients accurately.

Predicting poststroke dyskinesia with resting-state functional connectivity in the motor network

SIGNIFICANCE

Motor function evaluation is essential for poststroke dyskinesia rehabilitation. Neuroimaging techniques combined with machine learning help decode a patient's functional status. However, more research is needed to investigate how individual brain function information predicts the dyskinesia degree of stroke patients.

AIM

We investigated stroke patients' motor network reorganization and proposed a machine learning-based method to predict the patients' motor dysfunction.

APPROACH

Near-infrared spectroscopy (NIRS) was used to measure hemodynamic signals of the motor cortex in the resting state (RS) from 11 healthy subjects and 31 stroke patients, 15 with mild dyskinesia (Mild), and 16 with moderate-to-severe dyskinesia (MtS). The graph theory was used to analyze the motor network characteristics.

RESULTS

The small-world properties of the motor network were significantly different between groups: (1) clustering coefficient, local efficiency, and transitivity: MtS > Mild > Healthy and (2) global efficiency: MtS < Mild < Healthy. These four properties linearly correlated with patients' Fugl-Meyer Assessment scores. Using the small-world properties as features, we constructed support vector machine (SVM) models that classified the three groups of subjects with an accuracy of 85.7%.

CONCLUSIONS

Our results show that NIRS, RS functional connectivity, and SVM together constitute an effective method for assessing the poststroke dyskinesia degree at the individual level.

Deep Learning Paradigm for Cardiovascular Disease/Stroke Risk Stratification in Parkinson's Disease Affected by COVID-19: A Narrative Review

Background and Motivation: Parkinson's disease (PD) is one of the most serious, non-curable, and expensive to treat. Recently, machine learning (ML) has shown to be able to predict cardiovascular/stroke risk in PD patients. The presence of COVID-19 causes the ML systems to become severely non-linear and poses challenges in cardiovascular/stroke risk stratification. Further, due to comorbidity, sample size constraints, and poor scientific and clinical validation techniques, there have been no well-explained ML paradigms. Deep neural networks are powerful learning machines that generalize non-linear conditions. This study presents a novel investigation of deep learning (DL) solutions for CVD/stroke risk prediction in PD patients affected by the COVID-19 framework. Method: The PRISMA search strategy was used for the selection of 292 studies closely associated with the effect of PD on CVD risk in the COVID-19 framework. We study the hypothesis that PD in the presence of COVID-19 can cause more harm to the heart and brain than in non-COVID-19 conditions. COVID-19 lung damage severity can be used as a covariate during DL training model designs. We, therefore, propose a DL model for the estimation of, (i) COVID-19 lesions in computed tomography (CT) scans and (ii) combining the covariates of PD, COVID-19 lesions, office and laboratory arterial atherosclerotic image-based biomarkers, and medicine usage for the PD patients for the design of DL point-based models for CVD/stroke risk stratification. Results: We validated the feasibility of CVD/stroke risk stratification in PD patients in the presence of a COVID-19 environment and this was also verified. DL architectures like long short-term memory (LSTM), and recurrent neural network (RNN) were studied for CVD/stroke risk stratification showing powerful designs. Lastly, we examined the artificial intelligence bias and provided recommendations for early detection of CVD/stroke in PD patients in the presence of COVID-19. Conclusion: The DL is a very powerful tool for predicting CVD/stroke risk in PD patients affected by COVID-19.

Rapid triage for ischemic stroke: a machine learning-driven approach in the context of predictive, preventive and personalised medicine

BACKGROUND

Recognising the early signs of ischemic stroke (IS) in emergency settings has been challenging. Machine learning (ML), a robust tool for predictive, preventive and personalised medicine (PPPM/3PM), presents a possible solution for this issue and produces accurate predictions for real-time data processing.

METHODS

This investigation evaluated 4999 IS patients among a total of 10,476 adults included in the initial dataset, and 1076 IS subjects among 3935 participants in the external validation dataset. Six ML-based models for the prediction of IS were trained on the initial dataset of 10,476 participants (split participants into a training set [80%] and an internal validation set [20%]). Selected clinical laboratory features routinely assessed at admission were used to inform the models. Model performance was mainly evaluated by the area under the receiver operating characteristic (AUC) curve. Additional techniques-permutation feature importance (PFI), local interpretable model-agnostic explanations (LIME), and SHapley Additive exPlanations (SHAP)-were applied for explaining the black-box ML models.

RESULTS

Fifteen routine haematological and biochemical features were selected to establish ML-based models for the prediction of IS. The XGBoost-based model achieved the highest predictive performance, reaching AUCs of 0.91 (0.90-0.92) and 0.92 (0.91-0.93) in the internal and external datasets respectively. PFI globally revealed that demographic feature age, routine haematological parameters, haemoglobin and neutrophil count, and biochemical analytes total protein and high-density lipoprotein cholesterol were more influential on the model's prediction. LIME and SHAP showed similar local feature attribution explanations.

CONCLUSION

In the context of PPPM/3PM, we used the selected predictors obtained from the results of common blood tests to develop and validate ML-based models for the diagnosis of IS. The XGBoost-based model offers the most accurate prediction. By incorporating the individualised patient profile, this prediction tool is simple and quick to administer. This is promising to support subjective decision making in resource-limited settings or primary care, thereby shortening the time window for the treatment, and improving outcomes after IS.

SUPPLEMENTARY INFORMATION

The online version contains supplementary material available at 10.1007/s13167-022-00283-4.

Prediction of Hidden Coronary Artery Disease Using Machine Learning in Patients With Acute Ischemic Stroke

BACKGROUND AND OBJECTIVES

A machine learning technique for identifying hidden coronary artery disease (CAD) might be useful. We developed and validated machine learning models to predict patients with hidden CAD and assess long-term outcomes in patients with acute ischemic stroke.

METHODS

Multidetector coronary computed tomography was performed for patients without known history of CAD. Primary outcomes were defined as having any degree of CAD and having obstructive CAD (≥50% stenosis). Demographic variables, risk factors, laboratory results, Trial of ORG 10172 in Acute Stroke Treatment (TOAST) classification, NIH Stroke Scale score, blood pressure, and carotid artery stenosis were used to develop and validate machine learning models to predict CAD. Area under the receiver operating characteristic curves (AUC) was calculated for performance analysis, and Kaplan-Meier and Cox survival analyses of long-term outcomes were performed. Major adverse cardiovascular events (MACE) were defined as ischemic stroke, myocardial infarction, unstable angina, urgent coronary revascularization, and cardiovascular mortality.

RESULTS

Overall, 1,710 patients were included for the training dataset and 348 patients for the validation dataset. An Extreme Gradient Boosting model was developed to predict any degree of CAD, which showed an AUC of 0.763 (95% CI 0.711-0.814) on validation. A logistic regression model was used to predict obstructive CAD and had an AUC of 0.714 (95% CI 0.692-0.799). During the first 5 years of follow-up, MACE occurred more frequently when predicted of any CAD (P = 0.022) or obstructive CAD (P < 0.001). Cox proportional analysis showed that the hazard ratio of MACE was 1.5 (95% CI 1.1-2.2; P = 0.016) when predicted of any CAD, whereas it was 1.9 (95% CI 1.3-2.6; P < 0.001) for obstructive CAD.

DISCUSSION

We demonstrated that machine learning may help identify hidden CAD in patients with acute ischemic stroke. Long-term outcomes were also associated with prediction results.

CLASSIFICATION OF EVIDENCE

This study provides Class II evidence that in patients with acute ischemic stroke with CAD risk factors but no known history of CAD, a machine learning model predicts CAD on multidetector coronary computed tomography with an AUC of 0.763 (95% CI 0.711-0.814).

Use of artificial intelligence as an instrument of evaluation after stroke: a scoping review based on international classification of functioning, disability and health concept

Introduction: To understand the current practices in stroke evaluation, the main clinical decision support system and artificial intelligence (AI) technologies need to be understood to assist the therapist in obtaining better insights about impairments and level of activity and participation in persons with stroke during rehabilitation. Methods: This scoping review maps the use of AI for the functional evaluation of persons with stroke; the context involves any setting of rehabilitation. Data were extracted from CENTRAL, MEDLINE, EMBASE, LILACS, CINAHL, PEDRO Web of Science, IEEE Xplore, AAAI Publications, ACM Digital Library, MathSciNet, and arXiv up to January 2021. The data obtained from the literature review were summarized in a single dataset in which each reference paper was considered as an instance, and the study characteristics were considered as attributes. The attributes used for the multiple correspondence analysis were publication year, study type, sample size, age, stroke phase, stroke type, functional status, AI type, and AI function. Results: Forty-four studies were included. The analysis showed that spasticity analysis based on ML techniques was used for the cases of stroke with moderate functional status. The techniques of deep learning and pressure sensors were used for gait analysis. Machine learning techniques and algorithms were used for upper limb and reaching analyses. The inertial measurement unit technique was applied in studies where the functional status was between mild and severe. The fuzzy logic technique was used for activity classifiers. Conclusion: The prevailing research themes demonstrated the growing utility of AI algorithms for stroke evaluation.

Automatic Grading of Stroke Symptoms for Rapid Assessment Using Optimized Machine Learning and 4-Limb Kinematics: Clinical Validation Study

Background

Subtle abnormal motor signs are indications of serious neurological diseases. Although neurological deficits require fast initiation of treatment in a restricted time, it is difficult for nonspecialists to detect and objectively assess the symptoms. In the clinical environment, diagnoses and decisions are based on clinical grading methods, including the National Institutes of Health Stroke Scale (NIHSS) score or the Medical Research Council (MRC) score, which have been used to measure motor weakness. Objective grading in various environments is necessitated for consistent agreement among patients, caregivers, paramedics, and medical staff to facilitate rapid diagnoses and dispatches to appropriate medical centers.

Objective

In this study, we aimed to develop an autonomous grading system for stroke patients. We investigated the feasibility of our new system to assess motor weakness and grade NIHSS and MRC scores of 4 limbs, similar to the clinical examinations performed by medical staff.

Methods

We implemented an automatic grading system composed of a measuring unit with wearable sensors and a grading unit with optimized machine learning. Inertial sensors were attached to measure subtle weaknesses caused by paralysis of upper and lower limbs. We collected 60 instances of data with kinematic features of motor disorders from neurological examination and demographic information of stroke patients with NIHSS 0 or 1 and MRC 7, 8, or 9 grades in a stroke unit. Training data with 240 instances were generated using a synthetic minority oversampling technique to complement the imbalanced number of data between classes and low number of training data. We trained 2 representative machine learning algorithms, an ensemble and a support vector machine (SVM), to implement auto-NIHSS and auto-MRC grading. The optimized algorithms performed a 5-fold cross-validation and were searched by Bayes optimization in 30 trials. The trained model was tested with the 60 original hold-out instances for performance evaluation in accuracy, sensitivity, specificity, and area under the receiver operating characteristics curve (AUC).

Results

The proposed system can grade NIHSS scores with an accuracy of 83.3% and an AUC of 0.912 using an optimized ensemble algorithm, and it can grade with an accuracy of 80.0% and an AUC of 0.860 using an optimized SVM algorithm. The auto-MRC grading achieved an accuracy of 76.7% and a mean AUC of 0.870 in SVM classification and an accuracy of 78.3% and a mean AUC of 0.877 in ensemble classification.

Conclusions

The automatic grading system quantifies proximal weakness in real time and assesses symptoms through automatic grading. The pilot outcomes demonstrated the feasibility of remote monitoring of motor weakness caused by stroke. The system can facilitate consistent grading with instant assessment and expedite dispatches to appropriate hospitals and treatment initiation by sharing auto-MRC and auto-NIHSS scores between prehospital and hospital responses as an objective observation.

Accurate Prediction of Coronary Heart Disease for Patients With Hypertension From Electronic Health Records With Big Data and Machine-Learning Methods: Model Development and Performance Evaluation

BACKGROUND

Predictions of cardiovascular disease risks based on health records have long attracted broad research interests. Despite extensive efforts, the prediction accuracy has remained unsatisfactory. This raises the question as to whether the data insufficiency, statistical and machine-learning methods, or intrinsic noise have hindered the performance of previous approaches, and how these issues can be alleviated.

OBJECTIVE

Based on a large population of patients with hypertension in Shenzhen, China, we aimed to establish a high-precision coronary heart disease (CHD) prediction model through big data and machine-learning.

METHODS

Data from a large cohort of 42,676 patients with hypertension, including 20,156 patients with CHD onset, were investigated from electronic health records (EHRs) 1-3 years prior to CHD onset (for CHD-positive cases) or during a disease-free follow-up period of more than 3 years (for CHD-negative cases). The population was divided evenly into independent training and test datasets. Various machine-learning methods were adopted on the training set to achieve high-accuracy prediction models and the results were compared with traditional statistical methods and well-known risk scales. Comparison analyses were performed to investigate the effects of training sample size, factor sets, and modeling approaches on the prediction performance.

RESULTS

An ensemble method, XGBoost, achieved high accuracy in predicting 3-year CHD onset for the independent test dataset with an area under the receiver operating characteristic curve (AUC) value of 0.943. Comparison analysis showed that nonlinear models (K-nearest neighbor AUC 0.908, random forest AUC 0.938) outperform linear models (logistic regression AUC 0.865) on the same datasets, and machine-learning methods significantly surpassed traditional risk scales or fixed models (eg, Framingham cardiovascular disease risk models). Further analyses revealed that using time-dependent features obtained from multiple records, including both statistical variables and changing-trend variables, helped to improve the performance compared to using only static features. Subpopulation analysis showed that the impact of feature design had a more significant effect on model accuracy than the population size. Marginal effect analysis showed that both traditional and EHR factors exhibited highly nonlinear characteristics with respect to the risk scores.

CONCLUSIONS

We demonstrated that accurate risk prediction of CHD from EHRs is possible given a sufficiently large population of training data. Sophisticated machine-learning methods played an important role in tackling the heterogeneity and nonlinear nature of disease prediction. Moreover, accumulated EHR data over multiple time points provided additional features that were valuable for risk prediction. Our study highlights the importance of accumulating big data from EHRs for accurate disease predictions.

On the Role of Artificial Intelligence in Genomics to Enhance Precision Medicine

The complexity of orphan diseases, which are those that do not have an effective treatment, together with the high dimensionality of the genetic data used for their analysis and the high degree of uncertainty in the understanding of the mechanisms and genetic pathways which are involved in their development, motivate the use of advanced techniques of artificial intelligence and in-depth knowledge of molecular biology, which is crucial in order to find plausible solutions in drug design, including drug repositioning. Particularly, we show that the use of robust deep sampling methodologies of the altered genetics serves to obtain meaningful results and dramatically decreases the cost of research and development in drug design, influencing very positively the use of precision medicine and the outcomes in patients. The target-centric approach and the use of strong prior hypotheses that are not matched against reality (disease genetic data) are undoubtedly the cause of the high number of drug design failures and attrition rates. Sampling and prediction under uncertain conditions cannot be avoided in the development of precision medicine.

A Stroke Risk Detection: Improving Hybrid Feature Selection Method

Background

Stroke is one of the most common diseases that cause mortality. Detecting the risk of stroke for individuals is critical yet challenging because of a large number of risk factors for stroke.

Objective

This study aimed to address the limitation of ineffective feature selection in existing research on stroke risk detection. We have proposed a new feature selection method called weighting- and ranking-based hybrid feature selection (WRHFS) to select important risk factors for detecting ischemic stroke.

Methods

WRHFS integrates the strengths of various filter algorithms by following the principle of a wrapper approach. We employed a variety of filter-based feature selection models as the candidate set, including standard deviation, Pearson correlation coefficient, Fisher score, information gain, Relief algorithm, and chi-square test and used sensitivity, specificity, accuracy, and Youden index as performance metrics to evaluate the proposed method.

Results

This study chose 792 samples from the electronic records of 13,421 patients in a community hospital. Each sample included 28 features (24 blood test features and 4 demographic features). The results of evaluation showed that the proposed method selected 9 important features out of the original 28 features and significantly outperformed baseline methods. Their cumulative contribution was 0.51. The WRHFS method achieved a sensitivity of 82.7% (329/398), specificity of 80.4% (317/394), classification accuracy of 81.5% (645/792), and Youden index of 0.63 using only the top 9 features. We have also presented a chart for visualizing the risk of having ischemic strokes.

Conclusions

This study has proposed, developed, and evaluated a new feature selection method for identifying the most important features for building effective and parsimonious models for stroke risk detection. The findings of this research provide several novel research contributions and practical implications.

Defining and Predicting Pain Volatility in Users of the Manage My Pain App: Analysis Using Data Mining and Machine Learning Methods

Background

Measuring and predicting pain volatility (fluctuation or variability in pain scores over time) can help improve pain management. Perceptions of pain and its consequent disabling effects are often heightened under the conditions of greater uncertainty and unpredictability associated with pain volatility.

Objective

This study aimed to use data mining and machine learning methods to (1) define a new measure of pain volatility and (2) predict future pain volatility levels from users of the pain management app, Manage My Pain, based on demographic, clinical, and app use features.

Methods

Pain volatility was defined as the mean of absolute changes between 2 consecutive self-reported pain severity scores within the observation periods. The k-means clustering algorithm was applied to users’ pain volatility scores at the first and sixth month of app use to establish a threshold discriminating low from high volatility classes. Subsequently, we extracted 130 demographic, clinical, and app usage features from the first month of app use to predict these 2 volatility classes at the sixth month of app use. Prediction models were developed using 4 methods: (1) logistic regression with ridge estimators; (2) logistic regression with Least Absolute Shrinkage and Selection Operator; (3) Random Forests; and (4) Support Vector Machines. Overall prediction accuracy and accuracy for both classes were calculated to compare the performance of the prediction models. Training and testing were conducted using 5-fold cross validation. A class imbalance issue was addressed using a random subsampling of the training dataset. Users with at least five pain records in both the predictor and outcome periods (N=782 users) are included in the analysis.

Results

k-means clustering algorithm was applied to pain volatility scores to establish a threshold of 1.6 to differentiate between low and high volatility classes. After validating the threshold using random subsamples, 2 classes were created: low volatility (n=611) and high volatility (n=171). In this class-imbalanced dataset, all 4 prediction models achieved 78.1% (611/782) to 79.0% (618/782) in overall accuracy. However, all models have a prediction accuracy of less than 18.7% (32/171) for the high volatility class. After addressing the class imbalance issue using random subsampling, results improved across all models for the high volatility class to greater than 59.6% (102/171). The prediction model based on Random Forests performs the best as it consistently achieves approximately 70% accuracy for both classes across 3 random subsamples.

Conclusions

We propose a novel method for measuring pain volatility. Cluster analysis was applied to divide users into subsets of low and high volatility classes. These classes were then predicted at the sixth month of app use with an acceptable degree of accuracy using machine learning methods based on the features extracted from demographic, clinical, and app use information from the first month.

Disability Adjusted Life Years due to Ischaemic Stroke Preventable by Real-Time Stroke Detection-A Cost-Utility Analysis of Hypothetical Stroke Detection Devices

Background: Ischaemic stroke remains a significant contributor to permanent disability world-wide. Therapeutic interventions for acute ischaemic stroke (AIS) are available, but need to be administered early after symptom onset in order to be effective. Currently, one of the main factors responsible for poor clinical outcome is an unnecessary long time between symptom onset and arrival at a hospital (pre-hospital delay). In the future, technological devices with the capability of real-time detection of AIS may become available. The health economic implications of such devices have not been explored. Methods: We developed a novel probabilistic model to estimate the maximally allowable annual costs of different hypothetical real-time AIS detection devices in different populations given currently accepted willingness-to-pay thresholds. Distributions of model parameters were extracted from the literature. Effectiveness of the intervention was quantified as reduction in disability-adjusted life-years associated with faster access to thrombolysis and mechanical thrombectomy. Incremental costs were calculated from a societal perspective including acute treatment costs and long-term costs for nursing care, home help, and loss of production. The impact of individual model parameters was explored in one-way and multi-way sensitivity analyses. Results: The model yields significantly shorter prehospital delays and a higher proportion of acute ischaemic patients that fulfill the time-based eligibility criteria for thrombolysis or mechanical thrombectomy in the scenario with a real-time stroke detection device as compared to the control scenario. Depending on the sociodemographic and geographic characteristics of the study population and operating characteristics of the device, the maximally allowable annual cost for the device to operate in a cost-effective manner assuming a willingness-to-pay threshold of GBP 30.000 ranges from GBP 22.00 to GBP 9,952.00. Considering the results of multiway sensitivity analyses, the upper bound increases to GBP 29,449.10 in the subgroup of young patients with a very high annual risk of ischaemic stroke (50 years/20% annual risk). Conclusion: Data from probabilistic modeling suggest that real-time AIS detection devices can be expected to be cost-effective only for a small group of highly selected individuals.