19/06/2019 Almost all areas of health are becoming e-health - - PDF document

19/06/2019 1

Clinical decision support algorithms: how to ensure their safety and usefulness?

• Prof. Valérie D’Acremont, MD, PhD

Centre for primary care and public health, UNIL (Unisanté) Swiss Tropical and Public Health Institute, UNIBAS (SwissTPH) CISTM16, 9 June 2019

Almost all areas of health are becoming e-health…

Labrique et al, Global Health: Science and Practice 2013

Are electronic clinical decision support algorithms really new..?

1964

L.C. Payne, The role of the computer in refining diagnosis, The lancet

2013: Dr. Watson (IBM) “The times have passed when a single human mind could even pretend to know all that might be useful in aiding patients.”

Play Store

What is available on the market?

Symptom checker Isabel Babylon ada

Babylon at the heart of controversies…

8 complaints filed by GPs in UK. Babylon Internal study with 50 case scenarios: “Babylon do better diagnosis than human beings.” “Babylon technology is certified as a medical device.” (classe 1) “We are one of the safest primary care provider of UK.” Letter to the Lancet: “serious methodological problems” Some would like to see us fail and use anonymous and wrong allegations. Some even pretend to be physicians… The UK Care Quality Commission concludes that in some areas Babylon is not safe. (report censured by High Court). Babylon signs a contract with NHS. Letter to the BMJ: “Could Babylon please supply evidence?” 2.5 millions people in UK, Rwanda and Ireland are presently using Babylon…

51% of actual diagnoses were among the top 3 diagnoses provided by the algorithm

Semigran et al, BMJ 2015

1 2 3 4 5 6

19/06/2019 2

Ethics and challenges around digital health

WHO guideline digital interventions 2019

Ultimately, digital technologies are not ends in themselves; they are vital tools to promote health, keep the world safe, and serve the vulnerable.” “A key challenge is to ensure that all people enjoy the benefits

• f digital technologies for everyone.

We must make sure that innovation and technology helps to reduce the inequities in our world, instead of becoming another reason people are left behind. Countries must be guided by evidence to establish sustainable harmonized digital systems, not seduced by every new gadget. What type of algorithm for whom? Patient Specialized physician Front-line physician/clinician

• 1. Refer or admit?
• 2. Which lab test?
• 3. Meaning of the result

in the clinical context (specific treatment)?

You don’t have to go to a physician You don’t need to stay in bed but avoid physical efforts. Don’t stay in the sun and drink a lot. Take paracetamol. Dose of paracetamol SHOULD I GOTO A PHYSICIAN?

First step: Define target user and patient

Traveler or migrant upon return from the tropics with fever Primary care clinician Physician at hospital Pharmacist Community health worker Child 2 months – 5 years with history of fever

• r high temperature

Kristina Keitel et al, Plos Medicine 2017

12’124 articles

2nd step: Structured review of the literature 3rd step: studies to measure disease prevalence

Based on 25’743 biological tests Buss et al., in preparation

Tropical infections 45% Unknown 14% Acute febrile diarrhoea 15% Respiratory infections 13% Viral diseases 5% Other 3%

D’Acremont et al., NEJM 2014

Neurological infections 0.6% Mononucleosis 2% Noninfectious 2% Skin & soft tissue 3% Genitourinary 4%

Febrile Tanzanian children Febrile adult travellers

4th step: CART analyses to best combine clinical predictors

Sensitivity Specificity 46% 93% LR+ LR- 6.57 0.58 De Santis et al. Plos One 2017

7 8 9 10 11 12

19/06/2019 3

Erdman et al Plos One 2015

5th step: novel host biomarkers that predict disease

Radiological pneumonia in febrile Tanzanian children 28-day mortality in febrile Tanzanian adults

Richard-Greenblattet et al, CID 2019 ALMANACH, Clotilde Rambaud et al, Plos One 2013 (Adapted by Olga de Santis)

6th step: clinical decision support algorithm (CDSA)

      0100*+%=x011‐ %0110+x%1001x@01 101+00%@01

7th step: transform medical thinking into software coding MedAl-C: a software to transform your algorithm into an App The validation cycle of electronic clinical decision support algorithms

Clinical safety and efficacy Clinical effectiveness Impact, including

• n costs

Adaptation time & place, using generated data Clinical and epidemiological context Keitel & D’Acremont, Clin Microb & Infect 2018 I II III IV Validity and user-friendliness in the IT lab

13 14 15 16 17 18

19/06/2019 4

• Literature review
• Construction of decision charts with documentation
• quality of evidence
• strength of recommendation
• Reviewed by 15 experts at the CISTM7 (Insbruck,

2001)

• Publication in Journal of Travel Medicine 2003
• Update of guidelines in 2013 and in 2019 (new

assessment by experts planned)

Development of FeverTravel practice guidelines

Mueller et al, J. Travel Med. 2014

Online prospective study with 539 patient/clinician pairs

“No death was recorded and all complications could be attributed to the underlying illness rather than to adherence to guidelines.»

Vibert et al, in preparation Without App With App Exposures Symptoms General questions

Pilot study with GPs using FeverTravelApp on simulated patients

Number of questions proposed by the App

11/31 19/31

Algorithm to manage febrile children at primary care level (ePOCT)

Oximeter

Recommendation for treatment and/or admission

Hemoglobinometer Clinical data Malaria CRP PCT algorithm  diseases to be considered Glucometer

Randomized control trial of e-POCT

Kristina Keitel et al, Plos Medicine 2017

e-POCT Routine ALMANACH Cure rate at D3 and D7; 2nd hospitalisations and deaths by D30 3739 children 2 months - 5 years (9 facilities, Dar es Salaam)

Kristina Keitel et al, Plos Medicine 2017 20 40 60 80 100

Other Pneumonia Severe disease Routine ALMANACH ePOCT 100 98 96 94 92 90

96% 98% 95%

Potential impact of ePOCT in children in Tanzania: 1 million clinical failures averted per year

Impact of e-POCT implementation on cure rate

19 20 21 22 23 24

19/06/2019 5

Kristina Keitel et al, Plos Medicine 2017 20 40 60 80 100

Other Pneumonia Severe disease

30% 11% 95%

Routine ALMANACH ePOCT Potential impact of ePOCT in children in Tanzania: 28 million unnecessary antibiotics saved per year

Impact of e-POCT implementation on antibiotic prescriptions The validation cycle of electronic clinical decision algorithms

Kristina Keitel et al, Clin Microb & Infect 2018

DYNAMIC project

ePOCT clinical trial Clinical safety and efficacy Clinical effectiveness Impact, including

• n costs

Adaptation time & place, using generated data Clinical and epidemiological context I II III IV Validity and user-friendliness in the IT lab

The Tanzanian DYNAMIC project

e-POCT

ePOCT: - extended medical content

• new software
• full connection to biosensors and rapid tests

Validation:

• 70 health facilities
• 2 semi-urban districts

in Tanzania Beneficiaries: 500,000 sick children per year attending primary care facilities

10110100 1010101010

Dynamic algorithm: Through machine-learning and

• ptimization

Health system: Enhanced M&E, disease surveillance, epidemic detection Data sciences: High number and variability of data

Ecology and durability of smartphones/tablets implementation

Use a FAIRPHONE ! Don’t store useless data on long term !

Impact of algorithms beyond health

« If you record in the REC, you learn at the same time. If one day there is no tablet, you will still be able to correctly manage the child. » Accoucheuse ,Centre de Santé de Boulma

Bessat et al, BMC Public Health 2019

« Yes, it teaches us, as you cannot retain everything in your head. But with the REC, it reminds you at any time. At any time, you have it in front of you and it allows you to master. » Infirmier, Centre de Santé de Samba

It improves communication

« Now the clinicians ask us more questions

• n the child and touch him more. »

Président comité de gestion, village de Yako «On était dans les ténèbres. Maintenant, on est dans la lumière. » Chef du village de Yako Each technical innovation is doubled sided, not due to the good or bad way

• f using it, but due to the change in the distribution of power. It removes

power from some to give it to others, changing the reality for all. René Berger & Solange Ghernaouti-Hélie, ‘Technocivilisation, pour une philosophie du numérique’, 2010 It brings back pride and autonomy

IeDA project from Terre des hommes in Burkina Faso

30

25 26 27 28 29 30

19/06/2019 6

Name of CDSA Author/ developer POCTs used Algorithm content Clinical efficacy Clinical effectiveness Impact study Qualitative assessment Implemented iCCM based tools SL eCCM Imperial Col London mRDT CCM ‐ Ongoing ‐  Free on website e‐iCCM WHO, World Vision, Malaria Consortium ? ? ‐ Ongoing ? ? Niger and Mozambique eCCM D‐Tree, WEEMA mRDT CCM ‐    Ethiopia and Malawi IMCI based tools eIMCI D‐Tree, JHPIEGO, Harvard,Mariland mRDT, HIV RDT ? ‐  ‐ ‐ Zambia (large‐ scale planned) Neonatal IMNCI D‐Tree, Boston Children Hospital none Tanzanian IMNCI ‐  ‐ ‐ ‐ REC Terre des hommes mRDT Burkina Faso IMCI ‐ ‐ Ongoing  Burkina Faso (8 districts) Bangladesh digital IMCI MoH, ICDDRB mRDT Urine test Bangladesh IMCI ‐ ‐ ‐ Ongoing Bangladesh (3 sub‐districts) Kristina Keitel et al, Clin Microb & Infect 2018

Systematic review of CDSA for managing febrile children

Name of CDSA Author/ developer POCTs used Algorithm content Clinical efficacy Clinical effectiveness Impact study Qualitative assessment Implemented ALMANACH based tools e‐ALMANACH Swiss TPH mRDT Urine test Typhoid Published   ‐  Afghanistan Nigeria MSFeCARE MSF mRDT Urine test Oximeter Unpublished ‐ Unpublished ‐ Unpublished Central African R Niger, Tanzania, Mali Novel content based tools MEDSINC Think MD None, then mRDT ? ‐ Unpublished ‐ ‐ ‐ ePOCT SwissTPH mRDT, Hb Oximeter Glucomete CRP/PCT Published  ‐ ‐ ‐ ‐ Kristina Keitel et al, Clin Microb & Infect 2018

Systematic review of CDSA for managing febrile children

31 32