Department of Neurology Neurosciences Centre, and Clinical Epidemiology Unit, All India Institute of Medical Sciences, New Delhi Delhi, India
Knowledge of prognosis helps in:
Knowledge of prognosis helps in:
Care of individual patients:
For counselling patients (and/or their relatives) about their likely fate
Guiding our diagnostic and treatment decisions (e.g. withholding invasive tests or toxic treatments from those destined to have good prognosis without any treatments)
Comparing outcomes in certain groups of patients (treated in one hospital vs. another) to assess the quality of care
This section focuses on how to critically appraise articles that contain prognostic information that will be useful in counselling patients.
Q.1. Was There Any Selection Bias (Or Is the Study Sample Biased)?
Selection bias occurs when the patients under study are systematically different from the underlying population. This does not necessarily mean a conscious attempt on the part of the researchers to select a biased sample. Most often it occurs subconsciously because of the way study sample is assembled. Patients are included in a study because they both have a disease and are currently available (possibly because they are attending a clinic or a hospital).
1A. Why do we ask the question?
If the study sample is biased, then the results will be systematically different from what would have happened with unbiased sample. A disease begins in a member of a population. Let us call this a base population. The patients in a study include only those who are available for study some time after their disease began. For fatal conditions, such patients are the ones who are fortunate enough to have survived, and for diseases that remit, the patients are the ones who are unfortunate enough to have persistent disease. So, the biased sample may consist of only patients who survive and in whom the disease survives. This is why a term ‘survivor bias’ is sometimes used to describe this bias. Survivor bias may be a threat to validity of the conclusions of the study.
1B. How do we answer the question?
To detect selection bias in prognosis studies, look to see how the patients were assembled and what selection criteria were used. You ask: Is the criteria such that some patients from the base population would be missed because they recovered or died? If selection criteria are vague, the risk of biased sample is high. If it is clear, you consider the extent to which it will miss the serious or mild cases. For example, a hospital-based study of prognosis after stroke may have the criteria to include patients within the first month of onset. The patients who come in the third or fourth week after onset are those who survived the first 2 weeks and also did not recover. Many patients who developed stroke in the base population during the same week as these patients are not represented in the study. However, if the criteria include only those patients presenting within 72 h after onset may not suffer from the survivor bias to the same extent as above study. You should consider these aspects and decide whether there is high or low likelihood of bias in the sample selection.
1C. How do we interpret the answer?
If the likelihood of bias is high, then the results of the study are not useful to counsel the patients. However, there is perhaps no prognostic study, completely free from selection bias, and hence, you should be willing to accept a study with low likelihood of bias and assess further with the questions below.
Q.2. Did the Researchers Consider All Important Prognostic Factors?
Prognosis of most diseases is multifactorial. There are many prognostic factors that need to be considered while determining prognosis in an individual patient.
2A. Why do we ask the question?
If researchers did not consider certain important prognostic factors, the outcome rate of the study and prognostic prediction on its basis may not be valid. For example, if a study of brain haemorrhage reports 50 % mortality, but half of the patients are fully conscious and they all survive, whereas the other half are all comatose and they all die, the 50 % mortality rate may be valid for the group as a whole but not valid for a fully conscious patient or a comatose patient.
Researchers may consider the various prognostic factors in one of the two ways:
If there are few (one to three) important factors, they may form subgroups of patients according to the factors and provide outcome rates for each subgroup (referred to as risk stratification).
If there are many prognostic factors, use multivariable (regression) analysis to determine the most powerful predictors.
2B. How do we answer the question?
Based on your knowledge and experience of the condition, can you think of important prognostic factors that the researchers have ignored? If the answer is yes, the results may not be valid for applying to individual patients. Also look for results presented either as subgroups (or use of one of the multivariable analyses). Subgroup outcome rates may help you to get closer to your patient’s prognosis.
2C. How do we interpret the answer?
First of all, keep in mind that most of the time there is no way to be 100 % correct in predicting your individual patient’s prognosis. However, if outcome rates of subgroups are given, decide which subgroup your patient belongs to and use the outcome rate of that subgroup in communicating with your patient or his relatives. (Multivariable analysis is beyond the scope of this book.)
Q.3. Were Losses to Follow-Up Sufficiently Small?
Follow-up is a key factor in the validity and usefulness of a prognosis paper. There are two aspects of follow-up that need attention:
The losses to follow-up: The number of patients lost to follow-up threatens validity of the study.
Length of follow-up: Too short a follow-up may compromise the usefulness of the study. (Length of follow-up is discussed in the applicability section.)
Here we discuss the number lost to follow-up:
3A. Why do we ask this question?
If too many patients are lost to follow-up, the validity of the study may be threatened. This is so because the patients lost to follow-up tend to be systematically different from those who turn up for follow-up. Patients do not turn up for follow-up because they die, because they recover fully or because they are dissatisfied with the care and go to some other doctor or hospital. The larger the number of patients lost to follow-up relative to the number who suffered an adverse event, the greater is the threat to the study’s validity.
3B. How do we answer the question?
Look at the results section. Researchers describe (they should) how many patients they started with and how many were lost to follow-up. Sometimes, they do not mention anything about the lost to follow-up. Usually such studies have significant number of patients lost to follow-up. Validity of such studies remains open to question.
3C. How do we interpret the answer?