DECODE A computerized decision support system for the timely - PDF document

09/03/2016 DECODE A computerized decision support system for the timely identification of dementia Dr David Llewellyn University of Exeter Medical School “Britain unprepared for 'tsunami' of dementia patients” 1

09/03/2016 Cohort expectation of life at birth according to historic and projected mortality rates, for persons born from 1850 to 2050, England & Wales Source: Office for National Statistics Incidence rates of dementia in the United Kingdom compared with meta-analysis results in Europe and worldwide Nature Reviews Neuroscience 2007; 8: 233 – 9 2

09/03/2016 A sophisticated treatise on competing risk? Number of people with dementia in the UK, by level of severity and age group Source: Dementia UK: Update 3

09/03/2016 What is the probability that a randomly selected elderly patient will have dementia? What is the probability that a randomly selected elderly person has dementia? 40% 8.3% 6.5% 65% What is the probability that a randomly selected It depends on context... elderly patient will have dementia? What is the probability that a randomly selected elderly patient Torbay memory clinic, 2015 England, 1991 has dementia? 40% 8.3% 6.5% 65% England, 2011 English care home, 2011 4

09/03/2016 Substantial variation by age and time... ~670,000 people with dementia in 2011 (rather than the estimated 884,000) Lancet 2013; 382: 1405 – 12 Substantial variation by sex and location... (Estimated from Cambridgeshire, Nottingham and Newcastle) Lancet 2013; 382: 1405 – 12 5

09/03/2016 A common conventional (idealised) approach to case finding This is known to be inaccurate (unsystematic, subjective, complicated!) Limited accuracy (ignores most available information) Which cases are suitable for evaluation in primary care? 6

09/03/2016 Dementia case finding – a worked example ~6% of elderly people in England currently have dementia Population screening would identify most (not all) true cases True positives ( blue ): 5 False negatives ( red ): 1 MMSE: 88.3% sensitivity Lin, et al. (2013) Ann Int Med, 159, 9, 601-12. 7

09/03/2016 Population screening would identify most (not all) true cases True positives ( blue ): 5 False negatives ( red ): 1 If this triggered a 100% accurate diagnostic assessment(!) then undiagnosed dementia would drop to 12% MMSE: 88.3% sensitivity Lin, et al. (2013) Ann Int Med, 159, 9, 601-12. A huge number of ‘false positives’ without dementia though True negatives (black): 81 False negatives ( red ): 1 True positives ( blue ): 5 False positives ( green ): 13 MMSE: 86.2% specificity Lin, et al. (2013) Ann Int Med, 159, 9, 601-12. 8

09/03/2016 A huge number of ‘false positives’ without dementia though True negatives (black): 81 False negatives ( red ): 1 True positives ( blue ): 5 False positives ( green ): 13 Proportion of true positives = 29% (Worse than currently seen in memory clinics) MMSE: 86.2% specificity Lin, et al. (2013) Ann Int Med, 159, 9, 601-12. DECODE: Computerized decision support system for timely dementia identification Features: • Interactive • Intuitive • Evidence-based diagnostic ‘signatures’ • Embedded (ideally) and web versions • Incorporate guidelines and patient information • Promotes and enhances shared decision making 9

09/03/2016 Preliminary model development 685 older adults from the population-based Aging, Demographics and Memory Study (ADAMS) A wide range of primary care relevant potential predictors of dementia status evaluated (sociodemographics, medical history, medications, ADL/IADLs, cognition, self-report, informant report) Weighted stepwise multivariable logistic regression used to predict consensus DSM diagnosis of dementia status (yes/no) This is a standard (frequentist) model that ignores prior knowledge about the predictors Our full model will adopt a more complicated (Bayesian) approach Substantial improvement in dementia identification when weighted patient characteristics are used Misclassification rates are reduced from 22% to 6% 10

09/03/2016 DECODE will not replace clinical judgement! 11

09/03/2016 Thank you, my funders and colleagues Sir Halley Stewart Trust david.llewellyn@exeter.ac.uk 12