Evidence Based Practice Week 9

• Longitudinal studies follow patients over time to observe how factors influence later outcomes.
• Cross-sectional studies provide a "snapshot" of patients at a single time, showing how factors and outcomes align at that moment.
• Cohort studies track a group of individuals over time (often prospectively) to see who develops a particular outcome.
• Case-control studies start by identifying individuals based on whether they already have an outcome (cases) or not (controls), then look back (retrospectively) to determine potential causes.

Prognostic Research Studies

• Prognostic questions concern the impact of a disease or event on a patient’s long-term outcome.
• Prognostic questions can guide discharge planning, such as determining the likelihood of a patient returning home versus going to a rehabilitation facility.
• Prognostic assessments determine the odds an intervention will benefit a patient’s independent ambulation.
• Severity of a patient's problem, age, gender, home environment, and comorbidities can influence answers to prognostic questions.

Research Designs for Prognostic Studies

• Cohort studies provide data on the timing of outcome development within a group and help define possible causal factors.
• Case-control studies are conducted after an outcome has occurred with factors that contributed to the outcome studied in a group that has the outcome (case group) and compared to a group that does not have the outcome but is similar in other factors (control group).
• Cohort studies examine factors or risks for a particular outcome and observe the effect.
• In case-control studies, the event has already occurred, and the case group is compared to a control group that is similar but has not experienced the event.
• Cohort and case-control designs are commonly used in epidemiological research, where large groups are followed over long periods.
• Cross-sectional studies collect data at a specific point in time.
• Data in cross-sectional studies may be gathered over a week or longer, but during the time of the presenting problem.
• In cross-sectional studies with longer data-collection periods, researchers must ensure that no important changes occur that could seriously affect the outcome of interest.
• Prognostic studies show association, not causality.
• Randomized controlled trials or quasi-experimental designs are appropriate to test causal relations.
• Causal factors are factors that might be controlled or eliminated for a patient.

Determining Applicability and Quality

• The type of condition, age, comorbidities, and access to continued therapy influence prognosis.
• To determine applicability, search for a valid article that includes participants similar to your patient.
• The clinical usefulness of a prognosis study depends on both its applicability to the patient and the rigor of the study.
• Participants in prognostic studies are not randomly assigned to groups like in a randomized controlled trial.
• A group is identified that either has the outcome or may develop it.
• A sample must sufficiently represent the study event at a common point in the progression of the patients’ condition, such as the time of an injury.
• Prognostic studies typically have large samples due to the multifactorial nature of making prognostic statements.
• Accurate prognostic statements benefit from multiple contributing factors.
• With many factors affecting an outcome, the study must include enough participants with varying levels of the factors for analysis.
• Relevant factors and the common recruitment point determine the inclusion and exclusion criteria.
• In a longitudinal study, participants must not have the study outcome at the start.
• For example, a study on the effect of childhood diet on adolescent diabetes should only include participants who are diabetes-free at the outset.

Outcome Measures and Associated Factors

• Systematic and precise measurement time points (endpoints) should be clearly stated in the study.
• One or more measurement points may be included, but the time points should have relevance to the conditions and course of care.
• Outcome measures should be clearly defined, reliable, and valid.
• Reliability in other studies is insufficient; the people in the reported study must reliably perform the measures in the study sample under study conditions.
• Reliability within and between people is typically expressed by association (correlation) of repeated measurements.
• Both intra-rater and inter-rater reliability should have high values of association beyond chance.
• The type of correlation performed to assess rater reliability depends on the measurement scale used, as the scale determines the form of the data.
• Outcome measures must be explicitly stated, have validity in terms of their relevance to the study question, and should be reliable.
• The time points for outcome measures should be clear and consistent across participants.
• Factors hypothesized to be associated with the outcome of interest are identified in prognostic studies.
• A prognostic study is not likely to include all factors associated with a particular outcome, but it should identify the most important ones.

Study Process Essentials

• Members of the research study who conduct the measurements should be objective.
• Objectivity is best obtained if the "measurers" do not know the study purpose or the group status of the participants they are measuring.
• It is critical that the people measuring study outcome remain masked to as many features of the study and participant characteristics as possible.
• Relevant outcomes might take months or years to determine in a prognostic study.
• If different levels of the International Classification of Function, Disability and Health model are included in the outcome measures, the study should be sufficiently long enough to capture outcomes at all levels.
• The study may include a time point many months or even years after the study start date if subject participation in typical life roles is to be determined.
• Monitoring of participants should be specified.
• Participants may receive phone calls or visits during the study to check on compliance with study protocol or to motivate participants to continue participation in the study.
• Prognostic studies are typically observational studies.
• Factors measured are thought to be important contributors to the prognostic statements derived from the study, but other factors might also influence the study outcomes.
• Sufficient information should be collected on factors that might influence the outcomes.
• Subject attrition and causes for attrition should be stated.

Interpreting Prognostic Study Results

• Measures-of-association statistics are most appropriate for the analyses of prognostic studies.
• Association statistics are based on correlation.
• Common statistics for prognostic research in rehabilitation include correlations, linear and multiple regression, and logistic regression.
• In addition to these measures, statistics used in epidemiology are used for analysis in prognostic studies.
• Commonly used statistics are relative risk, expressed as odds ratios and risk ratios.
• Correlation: Measures the extent to which two variables are associated and how they change together.
• One factor does not necessarily cause the other factor to change but is associated with its change.
• Pattern of change/association between two variables over different values can be calculated as a correlation coefficient, represented as r.
• The association r varies between -1.0 and +1.0.
• A negative correlation of -1.0 means as one variable increases, the other decreases.
• A positive correlation of +1.0 means as one variable increases or decreases, the other varies in the same direction.
• The coefficient of determination, r^2, expresses the percentage of variance shared by two variables.
• Correlations are evaluated in terms of strength of association and the probability that chance accounts for the value (p-value).

Strength of Correlation (r-values)

• 0.00-0.25 = little or no correlation
• 0.26-0.49 = low strength
• 0.50-0.69 = moderate strength
• 0.70-0.89 = high strength
• 0.90-1.0 = very high strength
• Regression is a statistical analysis that builds on correlation statistics.
• It is used to make predictions from one or more variables about the outcome of interest.
• R^2 represents the results of a regression.
• The closer R^2 is to 1, the more robust the prediction.
• The regression equation will have a p-value, which expresses the amount that chance contributed to the prognostic equation.
• A significant p-value does not convey "meaningfulness"; a regression may be statistically significant but explain little of the outcome of interest.
• An R^2 value does not indicate a causal relationship between the selected variables in the equation and the outcome measure.
• Predictive statistics such as regression analyses have errors in the predictions.
• Confidence intervals may be included with the regression equation and should be included in your appraisal of the clinical implications of the results.
• A wide range in the confidence intervals reflects more error in the prediction.
• Simple Linear Regression: One variable (x) is used to predict the level of another variable (y).
• The assumption is that the two variable have a linear relationship.
• Multiple Regression: Predictions with multiple contributing variables are accomplished with the use of multiple-regression techniques.
• Multiple regression can be used to predict the incidence of stroke, where age and gender may be used in a prediction equation.
• Each variable contributes some amount to the outcome measure, as measured through a weighting process.
• The weighting process is expressed as beta-weights in the regression equation.
• The relationship among variables may be linear or non-linear, but in multiple regression, more than one variable is considered as contributing to the outcome of interest.
• Multiple regression is typically used when the outcome of interest is assessed with a continuous variable, even though some of the predictor variables used in the equations are categorical.
• Logistic Regression: Logistic regression is used when the outcome of interest is assessed with a categorical variable, typically dichotomous (two categories).
• Dichotomous outcomes can be derived from a continuous variable after systematically defining the outcome of interest as "present or absent" based on a cut-point on the continuous variable.
• Prognostic studies may use sensitivity, specificity, and likelihood ratios.
• Relative risk ratios (risk ratios and odds ratios) are common statistics used in case control studies.
• They are used to express the odds of the occurrence of a particular outcome between the case group and the control group.
• A single factor that has been identified for association with an outcome of interest is plotted against the actual outcome of interest for a sample of participants.
• If the factor and the outcome data are not in a dichotomous form, that is, in two categories, then two categories must be created.
• Categories created from continuous data such as a test score must be reduced to above or below a certain score.
• Risk Ratio (Also called relative risk) is a measure used to compare the risk of a particular event occurring in two different groups.
• Odds Ratio is a statistical measure used to determine the odds of an event occurring in one group relative to the odds of the same event occurring in another group.
• Risk ratios are typically calculated with a cohort design study in which the risk status is identified, and the cohort is followed to determine who develops the problem.
• Odds ratios are used with case control studies in which the problem is identified and then risk factors are investigated.
• Observational research designs are typically applied to prognostic questions.
• Statistical analyses of prognostic studies are based on measures of association and include correlation, regression, and estimates or relative risk.
• Prognostic studies can assist in making statements about expected outcomes for a patient or patient group and provide estimates of certain outcomes.