Professional Documents
Culture Documents
Destaye SA
December 2023
Objectives
At the end of this unit, you will be able to:
• Define prognosis
• Introduction
• Prognosis and prognostication
• Motives and aims of prognosis
• Approaches in prognostication
• Prognostic research
Practice-driven inquiry
• Often driven by the need to improve patient care and outcomes.
• Clinicians seek evidence-based answers to questions that directly impact their
ability to diagnose, treat, and prevent diseases in their patient
Provide a practically relevant answer that serves the practice
• Answers are expected to be directly applicable and relevant to clinical practice
• Findings should inform:
• Decision-making
• Clinical guidelines
• Healthcare policies Ultimately Improvements in patient care
Translational:
• Translates scientific discoveries into practical applications for patient care
• Risk assessment:
• It includes assessing the risk of specific events or complications related
to the health condition.
• This could involve estimating the risk of recurrence, complications, or
adverse effects.
February 27, 2024 Prognosis Research 14
• Time frame:
• Prognostication
• Can involve predictions for immediate outcomes or extend over
months or years.
• Individualized assessment:
• Prognostication is often individualized, taking into account the unique
characteristics of the patient, including:
o Age
o Overall health
o Comorbidities
o Response to treatment
• Communication:
• Clinicians need to convey predictions to patients to facilitate
shared decision-making
• Ethical considerations:
• Ethical considerations are often part of the prognostication process,
especially when making decisions about:
o Treatment intensity
o End-of-life care
o Respecting the patient's autonomy and preferences
• E.g: the 10-year survival rate for a woman between 50 and 70 years of age
with node-negative breast cancer with a tumor diameter of less than 10
millimeters is 93%
February 27, 2024 Prognosis Research 18
Motive and aim of prognosis
• Rooted in understanding and anticipating the future course of an individual's
health condition.
1. Informed decision-making:
• Prognosis is crucial:
• E.g. hip fracture can be accurately predicted from age, gender, height,
use of a walking aid, cigarette smoking, and body weight
• The aim of a prediction model in any medical field is not to take over the
job of the physician.
• It inherently longitudinal
• Involve time
Comprises :
• Nonclinical: age, gender
• Clinical: diagnosis, symptoms, signs, etiology, test results
• Has a major impact on the design of data collection and the design
of data analysis.
•
Outcome + Determinant + Domain + Setting
• Descriptive occurrence relations
o Joint space in patients with osteoarthritis of the knee (instead of pain, the
ability to walk, or quality of life)
• Blinding is not necessary for mortality or other outcomes that can be measured without
misclassification
• How data will be collected once the overall structure of the research is in
place
• The need to find the truth, and thus the need to never compromise validity,
is an essential starting point.
• Yet, for a given level of validity there may still be several options for the
collection of data.
• Sample:
• Only a sample from the study base
• Sometimes a case-control design (and thus a sampling rather than a
census approach) is used in prognostic research
Linear regression
• Continuous outcome
• e.g.: tumor size
• Discriminatory power assessed: Explained variance (R2)
Ordinal outcomes
• Model performance measures:
• Discrimination (e.g., c-index),
• Calibration (plots), and
• (re)classification measures
• Internal validation:
• If the number of potential predictors in multivariable regression modeling
is much larger than the number of outcomes or subjects, any fitted model
will result in overly optimistic predictive accuracy.
• Bootstrapping and cross-validation techniques can quantify the model’s
potential for:
• Overfitting
• Optimism in estimated model performance measures
• A shrinkage factor to adjust for this optimism
Analysis: Estimating added value
• Prognostic factors, tests, and biomarkers differ in
• Predictive accuracy
• Invasiveness, and
• Cost
• Tests or markers that are burdensome and costly to collect and measure should not be
evaluated on their individual predictive abilities
• but rather on the incremental predictive value beyond established, and easier-to-obtain,
predictors.
• Measures of discrimination such as the c-statistic are not able to detect small
improvements in model performance when a new marker is added to a model that
already includes important predictors.
• Recently, new metrics that estimate the added value of predictors have been proposed.
• These quantify the extent to which an extended model (with the addition of a
subsequent predictor or marker) improves the classification of participants with
and without the outcome compared with the basic model without that predictor.
• Correct reclassifications are shifts to a higher risk category in those who develop
the prognostic outcome and shifts to a lower risk category in those who do not.
• To circumvent this problem, a version of the NRI that does not require
stratification of the population into risk groups may be used.
• In contrast to the NRI, the IDI does not require subjectively predefined risk
thresholds.
• The IDI is the estimated improvement in the average sensitivity of the basic
model with the addition of the new predictor minus the estimated decrease in
the mean specificity, summarized over all possible risk thresholds.
Analysis: Consideration
• Missing values:
• completely case analysis may lead to biased results.
• Imputation (preferably MI) of missing values often yields less biased results.
• Continuous predictors:
• Should not be turned into dichotomies and linearity should not be assumed.
• Simple predictor transformation can be implemented to detect and model
nonlinearity, increasing the predictive accuracy of the prediction model.