Exploring the contextual assumptions, interventions and outcomes of digital advance care planning systems: A theory of change approach to understand implementation and evaluation.

BACKGROUND: Digital advance care planning systems are used internationally to document and share patients' wishes and preferences to inform care delivery. However, their use is impeded by a limited understanding of factors influencing implementation and evaluation. AIM: To develop mid-range programme theory to account for technological, infrastructure and human factor influences on digital advance care planning systems. DESIGN: Exploratory qualitative research design incorporating Theory of Change workshops that explored contextual assumptions affecting digital advance care planning in practice. A mid-range programme theory was developed through thematic framework analysis using the Non-adoption, Abandonment, Scale-up, Spread and Sustainability (NASSS) framework, generating a conceptual model depicting contextual assumptions, interventions and outcomes influencing implementation. PARTICIPANTS: A total of 38 participants (16 from London, 14 from West Yorkshire and 8 online) including patients, carers and health and care professionals (including those with commissioning responsibilities). RESULTS: A conceptual model was generated depicting five distinct components relating to digital advance care planning system use: (sociocultural, technical and structural prerequisites; recognition of the clinical need for conversation; having conversations and documenting decisions; accessing, actioning and amending; and using data to support evaluation, use and implementation). There were differences and uncertainty relating to what digital advance care planning systems are, who they are for and how they should be evaluated. CONCLUSIONS: Digital advance care planning lacks shared beliefs and practices, despite these being essential for complex technology implementation. Our mid-range programme theory can guide their further development and application by considering technological, infrastructure and human factor influences to optimise their implementation.

Deep learning of Parkinson's movement from video, without human-defined measures.

BACKGROUND: The core clinical sign of Parkinson's disease (PD) is bradykinesia, for which a standard test is finger tapping: the clinician observes a person repetitively tap finger and thumb together. That requires an expert eye, a scarce resource, and even experts show variability and inaccuracy. Existing applications of technology to finger tapping reduce the tapping signal to one-dimensional measures, with researcher-defined features derived from those measures. OBJECTIVES: (1) To apply a deep learning neural network directly to video of finger tapping, without human-defined measures/features, and determine classification accuracy for idiopathic PD versus controls. (2) To visualise the features learned by the model. METHODS: 152 smartphone videos of 10s finger tapping were collected from 40 people with PD and 37 controls. We down-sampled pixel dimensions and videos were split into 1 s clips. A 3D convolutional neural network was trained on these clips. RESULTS: For discriminating PD from controls, our model showed training accuracy 0.91, and test accuracy 0.69, with test precision 0.73, test recall 0.76 and test AUROC 0.76. We also report class activation maps for the five most predictive features. These show the spatial and temporal sections of video upon which the network focuses attention to make a prediction, including an apparent dropping thumb movement distinct for the PD group. CONCLUSIONS: A deep learning neural network can be applied directly to standard video of finger tapping, to distinguish PD from controls, without a requirement to extract a one-dimensional signal from the video, or pre-define tapping features.

Development and external validation of the eFalls tool: a multivariable prediction model for the risk of ED attendance or hospitalisation with a fall or fracture in older adults.

BACKGROUND: Falls are common in older adults and can devastate personal independence through injury such as fracture and fear of future falls. Methods to identify people for falls prevention interventions are currently limited, with high risks of bias in published prediction models. We have developed and externally validated the eFalls prediction model using routinely collected primary care electronic health records (EHR) to predict risk of emergency department attendance/hospitalisation with fall or fracture within 1 year. METHODS: Data comprised two independent, retrospective cohorts of adults aged ≥65 years: the population of Wales, from the Secure Anonymised Information Linkage Databank (model development); the population of Bradford and Airedale, England, from Connected Bradford (external validation). Predictors included electronic frailty index components, supplemented with variables informed by literature reviews and clinical expertise. Fall/fracture risk was modelled using multivariable logistic regression with a Least Absolute Shrinkage and Selection Operator penalty. Predictive performance was assessed through calibration, discrimination and clinical utility. Apparent, internal-external cross-validation and external validation performance were assessed across general practices and in clinically relevant subgroups. RESULTS: The model's discrimination performance (c-statistic) was 0.72 (95% confidence interval, CI: 0.68 to 0.76) on internal-external cross-validation and 0.82 (95% CI: 0.80 to 0.83) on external validation. Calibration was variable across practices, with some over-prediction in the validation population (calibration-in-the-large, -0.87; 95% CI: -0.96 to -0.78). Clinical utility on external validation was improved after recalibration. CONCLUSION: The eFalls prediction model shows good performance and could support proactive stratification for falls prevention services if appropriately embedded into primary care EHR systems.

Contactless hand tremor amplitude measurement using smartphones: development and pilot evaluation.

Background - Physiological tremor is defined as an involuntary and rhythmic shaking. Tremor of the hand is a key symptom of multiple neurological diseases, and its frequency and amplitude differs according to both disease type and disease progression. In routine clinical practice, tremor frequency and amplitude are assessed by expert rating using a 0 to 4 integer scale. Such ratings are subjective and have poor inter-rater reliability. There is thus a clinical need for a practical and accurate method for objectively assessing hand tremor.Objective - to develop a proof-of-principle method to measure hand tremor amplitude from smartphone videos.Methods - We created a computer vision pipeline that automatically extracts salient points on the hand and produces a 1-D time series of movement due to tremor, in pixels. Using the smartphones' depth measurement, we convert this measure into real distance units. We assessed the accuracy of the method using 60 videos of simulated tremor of different amplitudes from two healthy adults. Videos were taken at distances of 50, 75 and 100 cm between hand and camera. The participants had skin tone II and VI on the Fitzpatrick scale. We compared our method to a gold-standard measurement from a slide rule. Bland-Altman methods agreement analysis indicated a bias of 0.04 cm and 95% limits of agreement from -1.27 to 1.20 cm. Furthermore, we qualitatively observed that the method was robust to limited occlusion.Clinical relevance - We have demonstrated how tremor amplitude can be measured from smartphone videos. In conjunction with tremor frequency, this approach could be used to help diagnose and monitor neurological diseases.

Fixing the Leaky Pipe: How to Improve the Uptake of Patient-Reported Outcomes-Based Prognostic and Predictive Models in Cancer Clinical Practice.

PURPOSE: This discussion paper outlines challenges and proposes solutions for successfully implementing prediction models that incorporate patient-reported outcomes (PROs) in cancer practice. METHODS: We organized a full-day multidisciplinary meeting of people with expertise in cancer care delivery, PRO collection, PRO use in prediction modeling, computing, implementation, and decision science. The discussions presented here focused on identifying challenges to the development, implementation and use of prediction models incorporating PROs, and suggesting possible solutions. RESULTS: Specific challenges and solutions were identified across three broad areas. (1) Understanding decision making and implementation: necessitating multidisciplinary collaboration in the early stages and throughout; early stakeholder engagement to define the decision problem and ensure acceptability of PROs in prediction; understanding patient/clinician interpretation of PRO predictions and uncertainty to optimize prediction impact; striving for model integration into existing electronic health records; and early regulatory alignment. (2) Recognizing the limitations to PRO collection and their impact on prediction: incorporating validated, clinically important PROs to maximize model generalizability and clinical engagement; and minimizing missing PRO data (resulting from both structural digital exclusion and time-varying factors) to avoid exacerbating existing inequalities. (3) Statistical and modeling challenges: incorporating statistical methods to address missing data; ensuring predictive modeling recognizes complex causal relationships; and considering temporal and geographic recalibration so that model predictions reflect the relevant population. CONCLUSION: Developing and implementing PRO-based prediction models in cancer care requires extensive multidisciplinary working from the earliest stages, recognition of implementation challenges because of PRO collection and model presentation, and robust statistical methods to manage missing data, causality, and calibration. Prediction models incorporating PROs should be viewed as complex interventions, with their development and impact assessment carried out to reflect this.

Characteristics of good home-based end-of-life care: analysis of 5-year data from a nationwide mortality follow-back survey in England.

BACKGROUND: Recently, there has been an emphasis on providing good-quality end-of-life care; however, little is known about it and its determinants for patients living at home. AIM: To determine what characterises good-quality end-of-life care for patients living at home. DESIGN AND SETTING: An observational study using 5-year data from the National Survey of Bereaved People (Views of Informal Carers - Evaluation of Services [VOICES]) in England. METHOD: Analysis was based on data for 63 598 decedents, who were cared for at home in the last 3 months of life. Data were drawn from 110 311 completed mortality follow-back surveys of a stratified sample of 246 763 deaths registered in England between 2011 and 2015. Logistic regression analyses were used to identify independent variables associated with overall quality of end-of-life care and other indicators of end-of-life care quality. RESULTS: Patients who received good continuity of primary care (adjusted odds ratio [AOR] 2.03; 95% confidence interval [CI] = 2.01 to 2.06) and palliative care support (AOR 1.86; 95% CI = 1.84 to 1.89) experienced better overall quality of end-of-life care than those who did not, as perceived by relatives. Decedents who died from cancer (AOR 1.05; 95% CI = 1.03 to 1.06) or outside of hospital were more likely to receive good end-of-life care, as perceived by relatives. Being older, female (AOR 1.16; 95% CI = 1.15 to 1.17), from areas with least socioeconomic deprivation, and White (AOR 1.09; 95% CI = 1.06 to 1.12) were associated with better overall end-of-life care, as perceived by relatives. CONCLUSION: Better quality of end-of-life care was associated with good continuity of primary care, specialist palliative care support, and death outside of hospital. Disparities still exist for those from minority ethnic groups and those living in areas of socioeconomic deprivation. Future commissioning and initiatives must consider these variables to provide a more-equitable service.

Parkinsonian Hand or Clinician's Eye? Finger Tap Bradykinesia Interrater Reliability for 21 Movement Disorder Experts.

BACKGROUND: Bradykinesia is considered the fundamental motor feature of Parkinson's disease (PD). It is central to diagnosis, monitoring, and research outcomes. However, as a clinical sign determined purely by visual judgement, the reliability of humans to detect and measure bradykinesia remains unclear. OBJECTIVE: To establish interrater reliability for expert neurologists assessing bradykinesia during the finger tapping test, without cues from additional examination or history. METHODS: 21 movement disorder neurologists rated finger tapping bradykinesia, by Unified Parkinson's Disease Rating Scale (MDS-UPDRS) and Modified Bradykinesia Rating Scale (MBRS), in 133 videos of hands: 73 from 39 people with idiopathic PD, 60 from 30 healthy controls. Each neurologist rated 30 randomly-selected videos. 19 neurologists were also asked to judge whether the hand was PD or control. We calculated intraclass correlation coefficients (ICC) for absolute agreement and consistency of MDS-UPDRS ratings, using standard linear and cumulative linked mixed models. RESULTS: There was only moderate agreement for finger tapping MDS-UPDRS between neurologists, ICC 0.53 (standard linear model) and 0.65 (cumulative linked mixed model). Among control videos, 53% were rated > 0 by MDS-UPDRS, and 24% were rated as bradykinesia by MBRS subscore combination. Neurologists correctly identified PD/control status in 70% of videos, without strictly following bradykinesia presence/absence. CONCLUSION: Even experts show considerable disagreement about the level of bradykinesia on finger tapping, and frequently see bradykinesia in the hands of those without neurological disease. Bradykinesia is to some extent a phenomenon in the eye of the clinician rather than simply the hand of the person with PD.

Associations with Non-Persistence with Intra-Vitreal Therapy for Neovascular Age-Related Macular Degeneration at 24 Months.

AIMS: The aim of the study was to investigate non-persistence with treatment for neovascular age-related macular degeneration (NvAMD) before day 720 (24 months) after initiation, explore associations with baseline characteristics and variation between sites. METHODS: Anonymised demographic and clinical data were extracted from electronic medical records at treating National Health Service (NHS) Trusts for NvAMD eyes starting intra-vitreal therapy from 2017 to 2018. Time to non-persistence with treatment, defined as no recorded attendance for either monitoring or treatment for a period ≥6 months, was visualised with a Kaplan-Meier survival plot. Associations with treatment non-persistence were investigated using a Cox proportional hazards model. RESULTS: Analysis included 7,970 eyes of 7,112 patients treated at 13 NHS trusts. Censoring deaths and those eyes in which treatment was stopped permanently, the Kaplan-Meier analyses demonstrated survival figures of 77.7% for persistence with treatment to day 360 and 71.8% to day 720. Hazard ratios for non-persistence with treatment were reduced at 10 sites, relative to the reference, with first-treated eye status and with baseline acuity worse than or equal to LogMAR 1.0. Hazard ratios increased with younger age, in the presence of other ocular co-morbidities and with baseline acuity better than or equal to LogMAR 0.5. After an episode of non-persistence, visual acuity decreased by at least 0.1 and 0.3 LogMAR in 39% and 18% of eyes, respectively. CONCLUSIONS: Non-persistence with treatment was common, especially in the first year of treatment, and was often associated with a decrease in visual acuity. Treatment site, baseline visual acuity, and age were the strongest predictors of treatment non-persistence before day 720. Understanding and addressing reasons for non-persistence are important to ensure that effective but expensive treatments are used cost-effectively and to maintain acuity. Variation in non-persistence between sites, even after adjustment for other variables, suggests that local factors in treatment provision may be particularly important.

A novel temporal generative adversarial network for electrocardiography anomaly detection.

Cardiac abnormality detection from Electrocardiogram (ECG) signals is a common task for cardiologists. To facilitate efficient and objective detection, automated ECG classification by using deep learning based methods have been developed in recent years. Despite their impressive performance, these methods perform poorly when presented with cardiac abnormalities that are not well represented, or absent, in the training data. To this end, we propose a novel one-class classification based ECG anomaly detection generative adversarial network (GAN). Specifically, we embedded a Bi-directional Long-Short Term Memory (Bi-LSTM) layer into a GAN architecture and used a mini-batch discrimination training strategy in the discriminator to synthesis ECG signals. Our method generates samples to match the data distribution from normal signals of healthy group so that a generalised anomaly detector can be built reliably. The experimental results demonstrate our method outperforms several state-of-the-art semi-supervised learning based ECG anomaly detection algorithms and robustly detects the unknown anomaly class in the MIT-BIH arrhythmia database. Experiments show that our method achieves the accuracy of 95.5% and AUC of 95.9% which outperforms the most competitive baseline by 0.7% and 1.7% respectively. Our method may prove to be a helpful diagnostic method for helping cardiologists identify arrhythmias.

To the emergency room and back again: Circular healthcare pathways for acute functional neurological disorders.

BACKGROUND AND OBJECTIVES: Studies of Functional Neurological Disorders (FND) are usually outpatient-based. To inform service development, we aimed to describe patient pathways through healthcare events, and factors affecting risk of emergency department (ED) reattendance, for people presenting acutely with FND. METHODS: Acute neurology/stroke teams at a UK city hospital were contacted regularly over 8 months to log FND referrals. Electronic documentation was then reviewed for hospital healthcare events over the preceding 8 years. Patient pathways through healthcare events over time were mapped, and mixed effects logistic regression was performed for risk of ED reattendance within 1 year. RESULTS: In 8 months, 212 patients presented acutely with an initial referral suggesting FND. 20% had subsequent alternative diagnoses, but 162 patients were classified from documentation review as possible (17%), probable (28%) or definite (55%) FND. In the preceding 8 years, these 162 patients had 563 ED attendances and 1693 inpatient nights with functional symptoms, but only 26% were referred for psychological therapy, only 66% had a documented diagnosis, and care pathways looped around ED. Three better practice pathway steps were each associated with lower risk of subsequent ED reattendance: documented FND diagnosis (OR = 0.32, p = 0.004), referral to clinical psychology (OR = 0.35, p = 0.04) and outpatient neurology follow-up (OR = 0.25, p < 0.001). CONCLUSION: People that present acutely to a UK city hospital with FND tend to follow looping pathways through hospital healthcare events, centred around ED, with low rates of documented diagnosis and referral for psychological therapy. When better practice occurs, it is associated with lower risk of ED reattendance.

Risk scores in anaesthesia: the future is hard to predict.

External validation helps to assess whether a given risk prediction model will perform well in a target population. Validation is an important step in maintaining the utility of risk prediction models, as their ability to provide reliable risk estimates will deteriorate over time (calibration drift).

Suboptimal Dosing of ß-Blockers in Chronic Heart Failure: A Missed Opportunity?

BACKGROUND: The evidence base for the benefits of ß-blockers in heart failure with reduced ejection fraction (HFrEF) suggests that higher doses are associated with better outcomes. OBJECTIVES: The aim of this study was to report the proportion of patients receiving optimized doses of ß-blockers, outcomes, and factors associated with suboptimal dosing. METHODS: This was a prospective cohort study of 390 patients with HFrEF undergoing clinical and echocardiography assessment at baseline and at 1 year. RESULTS: Two hundred thirty-seven patients (61%) were receiving optimized doses (≥5-mg/d bisoprolol equivalent), 72 (18%) could not be up-titrated (because of heart rate < 60 beats/min or systolic blood pressure <100 mm Hg), and the remaining 81 (21%) should have been. Survival was similarly reduced in those who could not and should have been receiving 5 mg/d or greater, and patient factors did not explain the failure to attain optimized dosing. CONCLUSIONS: Many patients with HFrEF are not receiving optimal dosing of ß-blockers, and in around half, there was no clear contraindication in terms of heart rate or blood pressure.

Accuracy of Smartphone Video for Contactless Measurement of Hand Tremor Frequency.

BACKGROUND: Computer vision can measure movement from video without the time and access limitations of hospital accelerometry/electromyography or the requirement to hold or strap a smartphone accelerometer. OBJECTIVE: To compare computer vision measurement of hand tremor frequency from smartphone video with a gold standard measure accelerometer. METHODS: A total of 37 smartphone videos of hands, at rest and in posture, were recorded from 15 participants with tremor diagnoses (9 Parkinson's disease, 5 essential tremor, 1 functional tremor). Video pixel movement was measured using the computing technique of optical flow, with contemporaneous accelerometer recording. Fast Fourier transform and Bland-Altman analysis were applied. Tremor amplitude was scored by 2 clinicians. RESULTS: Bland-Altman analysis of dominant tremor frequency from smartphone video compared with accelerometer showed excellent agreement: 95% limits of agreement -0.38 Hz to +0.35 Hz. In 36 of 37 videos (97%), there was <0.5 Hz difference between computer vision and accelerometer measurement. There was no significant correlation between the level of agreement and tremor amplitude. CONCLUSION: The study suggests a potential new, contactless point-and-press measure of tremor frequency within standard clinical settings, research studies, or telemedicine.

Evidence of objective sleep impairment in nonepileptic attack disorder: A naturalistic prospective controlled study using actigraphy and daily sleep diaries over six nights.

Poor sleep is reported by many with nonepileptic attack disorder (NEAD) with correlations evident between self-reported sleep quality and mood and functional impairment. However, it is contended that self-reported sleep impairment in NEAD is a subjective phenomenon, which represents a general tendency to over-report symptoms or misinterpret bodily states in those with NEAD. The present study was therefore designed to investigate the extent of subjective and objective sleep impairments in those with NEAD. Over six nights we prospectively recorded comparable nightly objective (actigraphy) and subjective (consensus sleep diary) sleep parameters in a sample of 17 people with NEAD, and an age- and gender-matched normative control group (N = 20). Participants recorded daily measures of attacks, dissociation, and mood. Alongside higher subjective sleep impairment, the NEAD group had significantly worse objective sleep on several metrics compared to the normative controls, characterized by disrupted sleep (frequent awakenings and wake after sleep onset, low efficiency). Exploratory analyses using mixed effects models showed that attacks were more likely to occur on days preceded by longer, more restful sleep. This study, which had good ecological validity, evidences the presence of objective sleep impairment in NEAD, suggesting that in patient reports of problems with sleep should be given careful consideration in clinical practice.

Advanced care planning during the COVID-19 pandemic: ceiling of care decisions and their implications for observational data.

BACKGROUND: Observational studies investigating risk factors in coronavirus disease 2019 (COVID-19) have not considered the confounding effects of advanced care planning, such that a valid picture of risk for elderly, frail and multi-morbid patients is unknown. We aimed to report ceiling of care and cardiopulmonary resuscitation (CPR) decisions and their association with demographic and clinical characteristics as well as outcomes during the COVID-19 pandemic. METHODS: Retrospective, observational study conducted between 5th March and 7th May 2020 of all hospitalised patients with COVID-19. Ceiling of care and CPR decisions were documented using the Recommended Summary Plan for Emergency Care and Treatment (ReSPECT) process. Unadjusted and multivariable regression analyses were used to determine factors associated with ceiling of care decisions and death during hospitalisation. RESULTS: A total of 485 patients were included, of whom 409 (84·3%) had a documented ceiling of care; level one for 208 (50·9%), level two for 75 (18·3%) and level three for 126 (30·8%). CPR decisions were documented for 451 (93·0%) of whom 336 (74·5%) were 'not for resuscitation'. Advanced age, frailty, White-European ethnicity, a diagnosis of any co-morbidity and receipt of cardiovascular medications were associated with ceiling of care decisions. In a multivariable model only advanced age (odds 0·89, 0·86-0·93 p < 0·001), frailty (odds 0·48, 0·38-0·60, p < 0·001) and the cumulative number of co-morbidities (odds 0·72, 0·52-1·0, p = 0·048) were independently associated. Death during hospitalisation was independently associated with age, frailty and requirement for level two or three care. CONCLUSION: Ceiling of care decisions were made for the majority of patients during the COVID-19 pandemic, broadly in line with known predictors of poor outcomes in COVID-19, but with a focus on co-morbidities suggesting ICU admission might not be a reliable end-point for observational studies where advanced care planning is routine.

Association of heart failure and its comorbidities with loss of life expectancy.

OBJECTIVE: Estimating survival can aid care planning, but the use of absolute survival projections can be challenging for patients and clinicians to contextualise. We aimed to define how heart failure and its major comorbidities contribute to loss of actuarially predicted life expectancy. METHODS: We conducted an observational cohort study of 1794 adults with stable chronic heart failure and reduced left ventricular ejection fraction, recruited from cardiology outpatient departments of four UK hospitals. Data from an 11-year maximum (5-year median) follow-up period (999 deaths) were used to define how heart failure and its major comorbidities impact on survival, relative to an age-sex matched control UK population, using a relative survival framework. RESULTS: After 10 years, mortality in the reference control population was 29%. In people with heart failure, this increased by an additional 37% (95% CI 34% to 40%), equating to an additional 2.2 years of lost life or a 2.4-fold (2.2-2.5) excess loss of life. This excess was greater in men than women (2.4 years (2.2-2.7) vs 1.6 years (1.2-2.0); p<0.001). In patients without major comorbidity, men still experienced excess loss of life, while women experienced less and were non-significantly different from the reference population (1 year (0.6-1.5) vs 0.4 years (-0.3 to 1); p<0.001). Accrual of comorbidity was associated with substantial increases in excess lost life, particularly for diabetes, chronic kidney and lung disease. CONCLUSIONS: Comorbidity accounts for the majority of lost life expectancy in people with heart failure. Women, but not men, without comorbidity experience survival close to reference controls.

Seeing the unseen: Could Eulerian video magnification aid clinician detection of subclinical Parkinson's tremor?

INTRODUCTION: Eulerian magnification amplifies very small movements in video, revealing otherwise invisible motion. This raises the possibility that it could enable clinician visualisation of subclinical tremor using a standard camera. We tested whether Eulerian magnification of apparently atremulous hands reveals a Parkinsonian tremor more frequently in Parkinson's than in controls. METHOD: We applied Eulerian magnification to smartphone video of 48 hands that appeared atremulous during recording (22 hands from 11 control participants, 26 hands from 17 idiopathic Parkinson's participants). Videos were rated for Parkinsonian tremor appearance (yes/no) before and after Eulerian magnification by three movement disorder specialist neurologists. RESULTS: The proportion of hands correctly classified as Parkinsonian or not by clinicians was significantly higher after Eulerian magnification (OR = 2.67; CI = [1.39, 5.17]; p < 0.003). Parkinsonian-appearance tremors were seen after magnification in a number of control hands, but the proportion was greater in the Parkinson's hands. CONCLUSION: Eulerian magnification slightly improves clinician ability to identify apparently atremulous hands as Parkinsonian. This suggests that some of the apparent tremor revealed may be subclinical Parkinson's (pathological) tremor, and Eulerian magnification may represent a first step towards contactless visualisation of such tremor. However, the technique also reveals apparent tremor in control hands. Therefore, our method needs additional elaboration and would not be of direct clinical use in its current iteration.

Supervised classification of bradykinesia in Parkinson's disease from smartphone videos.

BACKGROUND: Slowness of movement, known as bradykinesia, is the core clinical sign of Parkinson's and fundamental to its diagnosis. Clinicians commonly assess bradykinesia by making a visual judgement of the patient tapping finger and thumb together repetitively. However, inter-rater agreement of expert assessments has been shown to be only moderate, at best. AIM: We propose a low-cost, contactless system using smartphone videos to automatically determine the presence of bradykinesia. METHODS: We collected 70 videos of finger-tap assessments in a clinical setting (40 Parkinson's hands, 30 control hands). Two clinical experts in Parkinson's, blinded to the diagnosis, evaluated the videos to give a grade of bradykinesia severity between 0 and 4 using the Unified Pakinson's Disease Rating Scale (UPDRS). We developed a computer vision approach that identifies regions related to hand motion and extracts clinically-relevant features. Dimensionality reduction was undertaken using principal component analysis before input to classification models (Naïve Bayes, Logistic Regression, Support Vector Machine) to predict no/slight bradykinesia (UPDRS = 0-1) or mild/moderate/severe bradykinesia (UPDRS = 2-4), and presence or absence of Parkinson's diagnosis. RESULTS: A Support Vector Machine with radial basis function kernels predicted presence of mild/moderate/severe bradykinesia with an estimated test accuracy of 0.8. A Naïve Bayes model predicted the presence of Parkinson's disease with estimated test accuracy 0.67. CONCLUSION: The method described here presents an approach for predicting bradykinesia from videos of finger-tapping tests. The method is robust to lighting conditions and camera positioning. On a set of pilot data, accuracy of bradykinesia prediction is comparable to that recorded by blinded human experts.

Time series clustering to examine presence of decrement in Parkinson's finger-tapping bradykinesia.

Parkinson's disease is diagnosed based on expert clinical observation of movements. One important clinical feature is decrement, whereby the range of finger motion decreases over the course of the observation. This decrement has been assumed to be linear but has not been examined closely.We previously developed a method to extract a time series representation of a finger-tapping clinical test from 137 smart- phone video recordings. Here, we show how the signal can be processed to visualize archetypal progression of decrement. We use k-means with features derived from dynamic time warping to compare similarity of time series. To generate the archetypal time series corresponding to each cluster, we apply both a simple arithmetic mean, and dynamic time warping barycenter averaging to the time series belonging to each cluster.Visual inspection of the cluster-average time series showed two main trends. These corresponded well with participants with no bradykinesia and participants with severe bradykinesia. The visualizations support the concept that decrement tends to present as a linear decrease in range of motion over time.Clinical relevance- Our work visually presents the archetypal types of bradykinesia amplitude decrement, as seen in the Parkinson's finger-tapping test. We found two main patterns, one corresponding to no bradykinesia, and the other showing linear decrement over time.