A web application for follow-up of results from a mobile device test battery for Parkinson's disease patients.

This paper describes a web-based system for enabling remote monitoring of patients with Parkinson's disease (PD) and supporting clinicians in treating their patients. The system consists of a patient node for subjective and objective data collection based on a handheld computer, a service node for data storage and processing, and a web application for data presentation. Using statistical and machine learning methods, time series of raw data are summarized into scores for conceptual symptom dimensions and an "overall test score" providing a comprehensive profile of patient's health during a test period of about one week. The handheld unit was used quarterly or biannually by 65 patients with advanced PD for up to four years at nine clinics in Sweden. The IBM Computer System Usability Questionnaire was administered to assess nurses' satisfaction with the web application. Results showed that a majority of the nurses were quite satisfied with the usability although a sizeable minority were not. Our findings support that this system can become an efficient tool to easily access relevant symptom information from the home environment of PD patients.

A pharmacokinetic-pharmacodynamic model for duodenal levodopa infusion.

OBJECTIVE: The purpose of this work was to identify and estimate a population pharmacokinetic- pharmacodynamic model for duodenal infusion of a levodopa/carbidopa gel (Duodopa) to examine pharmacological properties of this treatment. METHODS: The modeling involved pooling data from 3 studies (on advanced Parkinson disease) and fixing some parameters to values found in literature. The first study involved 12 patients studied on 3 occasions each and was previously published. The second study involved 3 patients on 2 occasions. A bolus dose was given after a washout during night. Plasma samples and motor ratings (clinical assessment of motor function on a 7-point treatment response scale ranging from "very off" to "very hyperkinetic") were collected until the clinical effect returned to baseline. The third study involved 5 patients on 3 occasions receiving 5 different dose levels. Different structural models were evaluated using the nonlinear mixed-effects modeling program NONMEM VI. Population mean parameter values, and interindividual, interoccasion, and residual variabilities were estimated. RESULTS: Absorption of the levodopa/carbidopa gel can be adequately described with first-order absorption with bioavailability and lag time. Estimated population parameter values were a mean absorption time of 28.5 minutes, a lag time of 2.9 minutes, and a bioavailability of 88%. The pharmacodynamic model for motor ratings had the following population values: a half-life of effect delay of 21 minutes, a concentration at 50% effect of 1.55 mg/L, an Emax of 2.39 U on the treatment response scale, and a sigmoidicity of the Emax function of 11.6. CONCLUSIONS: For the typical unmedicated subject, it will take 51.4 minutes until the peak levodopa effect is reached after a bolus dose. This delay is, like the magnitude of the effect, highly variable in this patient group. The residual error magnitudes of 20% for levodopa concentrations and 0.92 U (SD) for motor ratings indicate that the models developed provide predictions of a relevant quality. The developed model may be a first step toward model-guided treatment individualization of duodenal infusion of levodopa.

A new computer method for assessing drawing impairment in Parkinson's disease.

A test battery, consisting of self-assessments and motor tests (tapping and spiral drawing tasks) was used on 9482 test occasions by 62 patients with advanced Parkinson's disease (PD) in a telemedicine setting. On each test occasion, three Archimedes spirals were traced. A new computer method, using wavelet transforms and principal component analysis processed the spiral drawings to generate a spiral score. In a web interface, two PD specialists rated drawing impairment in spiral drawings from three random test occasions per patient, using a modification of the Bain & Findley 10-category scale. A standardised manual rating was defined as the mean of the two raters' assessments. Bland-Altman analysis was used to evaluate agreement between the spiral score and the standardised manual rating. Another selection of spiral drawings was used to estimate the Spearman rank correlations between the raters (r=0.87), and between the mean rating and the spiral score (r=0.89). The 95% confidence interval for the method's prediction errors was +/-1.5 scale units, which was similar to the differences between the human raters. In conclusion, the method could assess PD-related drawing impairments well comparable to trained raters.

A home environment test battery for status assessment in patients with advanced Parkinson's disease.

A test battery for assessing patient state in advanced Parkinson's disease, consisting of self-assessments and motor tests, was constructed and implemented on a hand computer with touch screen in a telemedicine setting. The aim of this work was to construct an assessment device, applicable during motor fluctuations in the patient's home environment. Selection of self-assessment questions was based on questions from an e-diary, previously used in a clinical trial. Both un-cued and cued tapping tests and spiral drawing tests were designed for capturing upper limb stiffnes, slowness and involuntary movements. The patient interface gave an audible signal at scheduled response times and was locked otherwise. Data messages in an XML-format were sent from the hand unit to a central server for storage, processing and presentation. In tapping tests, speed and accuracy were calculated and in spiral tests, standard deviation of frequency filtered radial drawing velocity was calculated. An overall test score, combining repeated assessments of the different test items during a test period, was defined based on principal component analysis and linear regression. An evaluation with two pilot patients before and after receiving new types of treatments was performed. Compliance and usability was assessed in a clinical trial (65 patients with advanced Parkinson's disease) and correlations between different test items and internal consistency were investigated. The test battery could detect treatment effect in the two pilot patients, both in self-assessments, tapping tests' results and spiral scores. It had good patient compliance and acceptable usability according to nine nurses. Correlation analysis showed that tapping results provided different information as compared to diary responses. Internal consistency of the test battery was good and learning effects in the tapping tests were small.

Dynamic decision support graph--visualization of ANN-generated diagnostic indications of pathological conditions developing over time.

OBJECTIVES: A common objection to using artificial neural networks in clinical decision support systems is that the reasoning behind diagnostic indications cannot be sufficiently well explained. This paper presents a method for visualizing diagnostic indications generated from an artificial neural network-based decision support algorithm (ANN-algorithm) in conditions developing over time. METHODS: The main idea behind the method is first to calculate and graphically present the decision regions corresponding to the diagnostic indications given as output from the ANN-algorithm, in the space of two selected, clinically established 'display variables'. Secondly, the trajectory of time series measurement results of these, often biochemical markers, together with the respective 95% confidence intervals are superimposed on the decision regions. This will permit a nurse or clinician to grasp the diagnostic indication graphically at a glance. The indication is further presented in relation to clinical variables that the clinician is already familiar with, thus providing a sort of explanation. The predictive value of the indication is expressed by the proximity of the measurement result to the decision boundary, separating the decision regions, and by a numerically calculated individualized predictive value. RESULTS: The method is illustrated as applied to a previously published ANN-algorithm for the early ruling-in and ruling-out of acute myocardial infarction, using monitoring of measurement results of myoglobin and troponin-I in plasma. CONCLUSION: The method is appropriate when there is a limited number of clinically established variables, i.e. variables which the clinician is used to taking into account in clinical reasoning.

Artificial neural network algorithms for early diagnosis of acute myocardial infarction and prediction of infarct size in chest pain patients.

BACKGROUND: To prospectively validate artificial neural network (ANN)-algorithms for early diagnosis of myocardial infarction (AMI) and prediction of 'major infarct' size in patients with chest pain and without ECG changes diagnostic for AMI. METHODS: Results of early and frequent Stratus CS measurements of troponin I (TnI) and myoglobin in 310 patients were used to validate four prespecified ANN-algorithms with use of cross-validation techniques. Two separate biochemical criteria for diagnosis of AMI were applied: TnI > or = 0.1 microg/L within 24 h ('TnI 0.1 AMI') and TnI > or = 0.4 microg/L within 24 h ('TnI 0.4 AMI'). To be considered clinically useful, the ANN-indications of AMI had to achieve a predefined positive predictive value (PPV) > or = 78% and a negative predictive value (NPV) > or = 94% at 2 h after admission. 'Major infarct' size was defined by peak levels of CK-MB within 24 h. RESULTS: For the best performing ANN-algorithms, the PPV and NPV for the indication of 'TnI 0.1 AMI' were 87% (p=0.009) and 99% (p=0.0001) at 2 h, respectively. For the indication of 'TnI 0.4 AMI', the PPV and NPV were 90% (p=0.006) and 99% (p=0.0004), respectively. Another ANN-algorithm predicted 'major AMI' at 2 h with a sensitivity of 96% and a specificity of 78%. Corresponding PPV and NPV were 73% and 97%, respectively. CONCLUSIONS: Specially designed ANN-algorithms allow diagnosis of AMI within 2 h of monitoring. These algorithms also allow early prediction of 'major AMI' size and could thus, be used as a valuable instrument for rapid assessment of chest pain patients.