From: The concept of justifiable healthcare and how big data can help us to achieve it
Step | Tool/method | Epistemologic aim | Pitfalls/problems | Role for big data/AI |
---|---|---|---|---|
Step 1: Evidence based medicine | ||||
Generate hypothesis | Physiology Pathophysiology Anatomy Biology Pathogenesis | Description: Describe biological processes in detail Causal Inference: Enhance mechanistic understanding of basic biological processes and how they can be influenced | Information overload Transportability/generalizability Incorrect analytical approach | Pattern recognition Classification Clustering Searching and Aggregating all already existing knowledge Prediction of mode of action of different molecular structures as derived from existing knowledge Propose alternative structures based on existing molecular knowledge |
 | Epidemiology | Description: Describe population in detail for given conditions and outcomes: prevalence, incidence, risk factors, associated factors Prediction: Predict relevant outcomes based on a given set of covariates for a given population (association) or an individual from that population (risk prediction or prognosis) Causal Inference: Enhance understanding of causal effect association between a covariate and an outcome in a given population | Data quality Transportability/generalizability Incorrect analytical approach (remaining confounding and bias; incorrect causal inference) | Pattern recognition Classification Clustering Searching and Aggregating all existing knowledge Prognostication Trial emulation Trial simulation Causal inference techniques Dynamic Decision Problems |
Proof of concept | (small) phase 1 and 2 trials | Description: describe properties of population Prediction: Predict relevant outcomes based on a given set of covariates for a given population (association) or an individual from that population (risk prediction or prognosis) Causal inference: Establish (near) causal effect association between a covariate and an outcome in a given population Estimation of effect size of an intervention on a given outcome in a given patient population | Publication bias Framing Relevance of outcomes | Searching and Aggregating all existing knowledge toavoid unnecessary duplication or indicate gaps in knowledge Prediction of mode of action of different molecular structures as derived from existing knowledge Trial emulation Trial simulation Causal inference techniques Dynamic Decision Problems |
Efficacy | Randomized controlled trial | Causal Inference: Estimation of effect size of an intervention on a given outcome in a given patient population | Information overload Internal validity (classic bias) Publication bias Incorrect analytical approach Framing Relevance of outcomes Rare side effects not captured in RCTs Transportability/generalizability For static decision problems only | Searching and Aggregating all existing knowledge to avoid unnecessary duplication or indicate gaps in knowledge Causal inference techniques to determine transportability to other patient populations Dynamic Decision Problems |
 | Observational trial or registry | Description: Describe associations between intervention and outcome Describe effect modifiers of that association Describe safety and association with side effects Prediction: Describe prognosis and evolution of conditions, and the modifiers Causal Inference: Estimation of effect size of an intervention on a given outcome in a given patient population in settings where RCTs are not possible or not ethical | (Unmeasured) Confounding Publication bias Framing Relevance of outcomes Transportability/generalizability | Trial emulation to explore residual confounding Trial simulation |
Effectiveness level 1 | Systematic review of randomized controlled trials | Description: Identify and aggregate available evidence Judge quantity and quality of available evidence Causal Inference: Based on effect size estimators, mean changes in outcomes can be estimated at population level when the intervention is implemented Based on subgroup analysis, change in outcome for subgroups can be estimated when the intervention is implemented | Information overload Publication bias Rare side effects not captured in RCTs | Searching and Aggregating all existing knowledge Trial emulation to explore transportability/generalizibility Pragmatic trial approaches |
 | Systematic review of observational trials | Description: Identify and aggregate available evidence Judge quantity and quality of available evidence Describe associations between intervention and outcomes Describe safety and association with side effects |  | Searching and Aggregating all existing knowledge Trial emulation to explore confounding |
Step | Tool/method | Epistemologic aim | Pitfalls/problems | Big data |
---|---|---|---|---|
Step 2: Socio-economic evaluation | ||||
Health technology assessment | Incremental Cost Effectiveness Ratio | Estimate cost of the intervention per quality adjusted life year (QALY) as compared to an alternative intervention | Â | Use wearables and online data collection to estimate QoL impact (patient reported outcomes/experience measures) Searching and Aggregating all existing knowledge on this and alternative interventions Detailed description and granularity of target population to help estimate impact in true society (transportability/generalizability) |
 | Budget Impact | Estimate total cost on population level of the intervention |  | Detailed description and granularity of target population to help estimate impact in true society (transportability/generalizability) |
 | Opportunity Cost | Estimate impact of implementation of this intervention on other interventions in the same or other comparable domain |  | Detailed description and granularity of target population to help estimate impact in true society (transportability/generalizability) Searching and Aggregating all existing knowledge on this and alternative interventions |
Societal evaluation | Quality assessment/improvement | Description: Describe interventions and outcomes in different settings Describe effect modifiers and case-mix in different settings Prediction: Identify subpopulations where certain interventions have different outcome Causal Inference: Estimation of effect size of an intervention on a real world population and conditions | Data collection is resource heavy(cost/labor force) Missing data/Incomplete data Cherry picking | Automatic online registration of relevant parameters to allow quality assessment in a sustainable, reliable, complete and cost-effective way |
 | Pharmacovigilance (Side effects) | Describe safety and association with side effects | Rare side effects not captured in RCTs Side effects might be different in subpopulations not represented in RCTs | Detailed description and granularity of target population to help estimate impact in true society (transportability/generalizability) |
 | Patient reported outcome/experience measure (PROM/PREM) | Description: Describe interventions and patient-relevant outcomes in different settings Prediction: Identify subpopulations where certain interventions have different outcome Causal Inference: Estimationw of effect size of an intervention on a real world population and conditions | RCTs not representative for real world practice and not using patient relevant outcomes Differences in implementation might cause differences in outcomes | Automatic online registration of relevant parameters to allow quality assessment in a sustainable, reliable, complete and cost-effective way |