Eduard Kieser

Hydration sensor development for use on infants in underserved areas by Eduard Kieser, 10 November 2015 1 Background and Motivation Under 5 mortality by cause in resource poor settings 2 Background and Motivation • Diarrhoea and the associated dehydration is still a massive burden. • Resource poor settings are most affected. Under 5 mortality by cause in high resource settings 3 Background and Motivation How is dehydration measured in a clinical setting? • Using clinical indicators • Capillary refill time, skin recoil time, breathing tempo, general appearance and many more. • Using dehydration scales • WHO scale, CDS, Gorelick. • Pringle has shown that the scales perform poorly in underserved settings. • The subjective nature of the markers has been identified as a potential reason for this. 4 Our mission To develop and test 3 different non-invasive optical sensors that can improve the prospective hydration assessment of infants by minimizing the human error that is involved in measuring the clinical markers. 5 Design considerations • The final sensors • • • • • • Low-cost Portable Robust and reliable Unobtrusive User friendly Fit clinical workflow • The proof of concept devices • Capture the maximum amount of information • Ergonomic enough for use in a clinical setting • Implement secure data handling 6 The Capillary Refill Time sensor *Steiner, M.J. and Dewalt, D.A.: Is This Child Dehydrated ? The Rational Clinical Examination, vol. 291, no. 22, pp-, 2004. 7 The Capillary Refill Time sensor 8 The Skin Recoil Time sensor *Steiner, M.J. and Dewalt, D.A.: Is This Child Dehydrated ? The Rational Clinical Examination, vol. 291, no. 22, pp-, 2004. 9 The Skin Recoil Time sensor 10 The Thermal Profile sensor 11 The Thermal Profile sensor 12 The data extraction application • Stand alone interactive GUI application • Pre-processing • Marker extraction • Data handling 13 The infant study • Was conducted at Tygerberg Hospital Paediatric Emergency and Ambulatory Unit. • Patients were measured every four to six hours. • Post illness weight gain was used as the reference value. • Conventional clinical markers were also captured in order to compare the sensor performance to that of the clinical markers. 14 Patient summary Patient Age Admission Discharge Percent Length of Weight zGender Number (months) Weight (Kg) Weight (Kg) Dehydration Stay (days) score 1 9.6 m 10.25 10.42 2.1 0.9 1.265 2 9.3 f 6.46 7.3 11 2.6 -1.07 3 10.3 f 6.6 6.97 3.9 1.6 -1.73 4 5.2 f 5.1 5.85 12.8 0.9 -1.463 5 14 m 7.5 8.56 12.4 3.9 -1.509 6 12 f 9.16 9.39 2 1.1 0.356 7 5.7 m 6.14 6.43 -1.1 0.9 -1.77 8 5.7 m 6.05 6.18 2.9 0.9 -2.117 9 10.4 m 9.49 9.48 -0.2 1.7 0.206 10 29.8 m 11 12.75 13.7 1.9 -0.357 15 Raw results 16 Combining predictors Exploring data separability using dimensionality reduction • Linear discriminant analysis with all the continuous predictors. • LDA is used to project a high dimensional matrix to a lower dimensional sub space while maximizing class seperability. • Kernel density estimate of the first linear discriminant shows limited class seperability for the 5% dehydration threshold. (eigenvalues = [0.874, 7.34e-17, 4.48e-17, 6.72e-18, 0.0] ) 17 Combining predictors Finding the best predictors • Standardised ordinary least squares (OLS) linear regression was done with all the available predictors • Sensor measurements • Clinical assessments (doctors) • Routine assessments (nurses) • Standardised input • Larger predictor coefficient indicates better correlation 18 Combining predictors Building the model • Unstandardized OLS using the selected markers • Temperature profile • Skin recoil time 19 Sensor performance comparison for infant study 1 0.9 0.8 - Fusion CRT sensor SRT sensor TP sensor Sensitivity CDS Gorelick WHO CRT clinical SRT clinical Specificity 20 Conclusions • Basic operational feasibility of the 3 sensors (CRT, SRT and TP) was demonstrated. • Data obtained are not statistically significant due to the small sample size of the study. • The SRT and TP sensors yielded encouraging results which provide a basis for further exploration. 21 Thank you 22 23 24 25 Preliminary sensor validation Repeatability and measurement/noise ratio assessment on adults 26 27