Table 1.

Studies of prognosis

Research Purpose	Study Design	Analysis	Performance Measures	Model interpretation (From Fig. 2b multivariable regression model)	Application
Exploratory prognostic studies
2.1.1. Description
To describe the outcomes and course of people with a health condition. E.g., What is the course of recovery for adults with acute back pain (within 7 days of onset)?	Cohort (ideally an inception cohorta)	Descriptive statistics. For example, measure pain severity and function at pre-specified time intervals. Trajectory analysis can also be useful.	N/A	N/A	Understanding the course of a disease or exploring trajectories of recovery. May also indicate which outcomes could be tested for an association with candidate prognostic factors
2.1.2 Association
To identify candidate prognostic factors (prognostic markers /determinants). E.g., What factors are associated with disability in adults 12 months after onset of an episode of back pain?	Cohort (ideally an inception cohorta) and case-control studies.	Ideally, a multivariable model focusing on the strength of association between each candidate prognostic factor and an outcome.	Strength of association: the size of the beta-coefficient, odds / risk / hazard ratio, the width of the 95% confidence interval, and the statistical significance for each candidate prognostic factor	All three factors are associated with disability at 12 months: back pain duration (13.2, 95%CI 11.0, 15.5), baseline disability (0.29, 95%CI 0.25, .33), recovery expectations (− 3.2, 95%CI − 3.5, − 2.8). Note: the strength of association depends on other factors in model and are not directly comparable when prognostic factors are measured on different scales	Indicate which prognostic factors might be considered for use in predictive models and causal research
2.1.3 Prediction Model Development
To determine predictors (prognostic markers/determinants) of an outcome. What is the probability of an outcome? E.g., What predicts disability in adults 12 months after onset of an episode of back pain?	Inception cohorta, although sometimes a prevalence cohort is used if the intended clinical application of the model requires it	Multivariable model	Collective predictive ability of a set of predictors. Common measures of predictive ability include discrimination, calibration, R2.	Prediction model with the 3 predictors (back pain duration, baseline disability, and baseline recovery expectations) predicts disability at 12 months (adjusted R2 = 0.39)	Identification of chosen model is followed by the need for testing its external validity
Confirmatory prognostic studies
2.2.1 Prediction Model External Validation
To determine if the prediction model predicts well in external populations. E.g., What predicts disability in adults 12 months after onset of an episode of back pain?	Cohort (as above)	Apply coefficients for each predictor (from model development) to this new cohort	Model performs well in this independent cohort (similar to how it performed in development cohort). Common measures of model performance include model fit, discrimination, calibration and shrinkage.	N/A	Translate into clinical prediction/decision rules
2.2.3 Studies of causation
To determine if a candidate prognostic variable is a prognostic determinant (cause) of an outcome. E.g., Is recovery expectation a prognostic factor of disability 12 months after onset of an episode of back pain?	Inception cohorta	Test pre-specified hypothesis. Multivariable model. There are many research designs for different causal questions. One simple design is to determine whether an independent association exists between the potential prognostic determinant and an outcome, while controlling for potential confounders	Strength of association (effect estimate), its 95% confidence interval, and p-value in the presence of potential confounders	Recovery expectation is a prognostic determinant (cause) of disability at 12 months (−3.18, 95% CI −3.5, −2.8) independent of back pain duration and baseline disability.	Develop and test interventions targeted at the modifiable prognostic determinant. For example, to test whether improving patients’ expectations results in better outcomes
Clinical application
2.2.2 Clinical Prediction or Decision Rules
Clinical prediction rule: A version of the prediction model that has been simplified for clinical use. A tool used in the clinic that helps inform patients and clinicians about the probability of an outcome. Clinical decision rule: assists clinicians with decision-making and care pathways. E.g., A prediction rule indicating which people have a higher probability of responding well to a particular therapeutic intervention.	A before and after design		Feasibility, clinician and patient acceptance, estimates of likely effect on patient outcomes and/or health system outcomes		Determine whether effect should be subsequently tested in an intervention study.
To determine the impact of using a clinical prediction/decision rule on patient outcomes or cost-effectiveness of care. E.g., What is the impact of implementing the use of a clinical decision rule in adults with back pain?	Randomised controlled trial		Measures of impact: clinician adoption rates, clinician and patient acceptability, change in decision-making, improvement in patient, health system and economic outcomes		Recommend clinical prediction/decision rules for use in clinical practice.

^aInception cohort: participants are incepted at a uniform time (zero time), such as at the onset of a condition of interest or new episode of a condition of interest or onset of care-seeking, and are then followed over time for the development of outcome(s)