CHAPTER 8 Evidence about prognosis
After reading this chapter, you should be able to:
Let us consider a clinical scenario that will be useful for illustrating the concepts of evidence about prognosis that are the focus of this chapter.
Clinical scenario
Mrs Wilson is a 68-year-old woman with osteoarthritis affecting her knees and left hip. Her local doctor has referred her to an orthopaedic surgeon regarding her right knee. The pain in her right knee has been worsening for the past 6 months, making it difficult to manage stairs. The surgeon has recommended a total knee arthroplasty (knee replacement). Mrs Wilson also has mild chronic obstructive pulmonary disease. She takes no medication for this and has not smoked for almost 30 years. Mrs Wilson cares for her husband who has Parkinson’s disease. He recently fractured his arm in a fall at home, but has recovered well. Mrs Wilson has not decided whether to proceed with the total knee arthroplasty.
The clinical scenario above raises several questions about the future of the client’s condition. What change in pain and mobility can Mrs Wilson expect if she chooses not to undergo the surgery? If she decides not to have the operation now, for how long will she remain a suitable candidate for surgery? Assuming the surgery is performed, many more questions are raised. How long will it take for the immediate symptoms associated with the surgery to resolve? Will complications occur? How much improvement in mobility can she expect after the surgery? How long will it take to achieve this level of mobility? Will surgical revision of the joint replacement become necessary and, if so, when? These types of questions will be the focus of this chapter on prognosis.
Prognosis is about predicting the future—the future of a client’s condition. While it is impossible for anyone to predict the future with absolute certainty, we can use evidence from the past to make informed and sensible predictions about the future. These evidence-based predictions about the future can be useful in many ways. They can help reassure clients by removing some doubt about the future, especially if their expectations are unjustifiably pessimistic. Predictions about natural recovery can help you and your client to jointly decide whether interventions need to be considered at all. Sometimes an intervention is chosen that is typically applied only once, as with joint replacement or organ transplant. In such cases, the optimal time to apply the intervention can be determined by predictions about the rate of deterioration before it and the rate of recovery after it. Predictions about the average course of a particular condition can also be adjusted for individual clients. This adjustment is possible when other features about a client or the client’s health or management besides the primary diagnosis have been shown to affect outcomes.
This chapter will address the process of using prognostic evidence to make these predictions and incorporating them into clinical practice. We will start by defining the components of a structured clinical question about prognosis. Then we will see how to appraise the evidence to determine its likely validity. Subsequent sections of the chapter will review how to understand the results of a prognostic study, how to use the evidence to inform practice and how to explain prognostic information clearly to clients.
How to structure a prognostic question
You will recall from Chapter 2 that clinical questions can be structured using the PICO format: patient/problem, intervention/issue, comparison (if relevant) and outcomes. When our question was about the effect of intervention, the comparison was an important component. The effect of an intervention was always estimated by comparison against this component, even if it was a ‘no-intervention’ or usual care control. Questions about prognosis, instead, are questions about expected outcomes, not questions about what has caused those outcomes. Therefore, the comparison component is not used in questions about prognosis. Let us look at each of the remaining components in more detail.
Patient/problem
The patient/problem component can simply be specified as previously described, for example, ‘In clients with coronary heart disease …’, ‘Among children with epilepsy …’ or, from our scenario, ‘In adults with osteoarthritis of the knee …’. Sometimes, the prognosis for typical clients with the condition is quite different to the prognosis for clients with some extra characteristic. For example, the prognosis for clients with cystic fibrosis who become infected with the bacteria Burkholderia cepacia is worse than for those who do not.1 Characteristics that influence outcomes are known as prognostic factors. If you suspect that some characteristic of your client might be a prognostic factor, this can be incorporated into the patient/problem component. Let us assume for a moment that Mrs Wilson, the client in our scenario at the beginning of this chapter, is mildly obese. This may be a prognostic factor, so we could incorporate this into our clinical question: In adults with osteoarthritis of the knee who are obese …. In addition to comorbidities like obesity, prognostic factors can also relate to the severity of the condition, for example, ‘In clients with coronary heart disease (New York Heart Association Functional Class IV) …’. The New York Heart Association functional classification is a simple way of describing the extent of heart disease. It places clients in one of four categories based on the severity of their symptoms and how much they are limited during exercise. The history of the condition can also be a prognostic factor, for example, ‘Among children who have had their first epileptic seizure …’.
Intervention/issue
The next component of the question is the intervention/issue. If you are interested in the natural course of a condition, then you can simply add the term ‘untreated’ to your clinical question, for example, ‘In children with untreated nocturnal enuresis …’. This will remind you when you search for evidence that you are interested in prognostic evidence about untreated clients. It is logical to assume that a client’s prognosis may be affected by receiving an intervention, especially if the intervention has been shown to be effective. Therefore, a clinical question about prognosis should specify what intervention a client has received or is receiving for their condition. In fact, some questions are only relevant to a population that has received an intervention, such as in these two examples, ‘In clients undergoing surgical skin grafts for major burns, what is the risk of postoperative complications?’ and ‘Among clients who no longer stutter at the end of a course of intensive therapy, what is the probability that their stuttering will relapse in the next year?’
Outcomes
The last component of a clinical question about prognosis is outcomes. It is important to consider outcomes that will have the greatest impact on the client’s goals and priorities. The prognosis can also change over time. For example, among alcoholic women who are able to stop drinking alcohol and remain abstained from it, the average improvements in memory and psychomotor speed at 1 year are minimal, while by 4 years they have usually returned to within the normal range.2 Therefore, it is sometimes worthwhile adding a time frame to the outcome component of your clinical question.
Clinical scenario (continued): Structuring the clinical question
There are many prognostic questions that can be drawn from our scenario. Let us assume that Mrs Wilson has had further discussions with her orthopaedic surgeon and has now decided to go ahead with the surgery. Her primary concern is arranging care for her husband while she is incapacitated by the surgery. Mrs Wilson is keen to know how long it will take for her to regain the function in her knee, particularly her ability to walk independently. Mrs Wilson is eligible for 2 weeks of respite care and the Wilsons’ son is able to take 4 weeks off work to assist with the care of his father. From this scenario, a suitable prognostic question would be: In clients undergoing total knee arthroplasty for osteoarthritis, what improvement in walking ability would be expected after 6 weeks?
Clinical scenario (continued): Finding the evidence to answer your question
You start by looking for a prospective cohort study in PubMed Clinical Queries, filtering your search with the ‘prognosis’ and ‘narrow’ options selected and using the search terms: osteoarthritis AND (‘total knee’ OR TKA OR TKR) AND walking ability. You have included TKA and TKR in your search terms as total knee arthroplasty is sometimes abbreviated as TKA and is sometimes referred to as total knee replacement or its abbreviation, TKR. This search results in nine articles. A quick scan of the titles confirms that several of the articles are probably relevant. One of these is very close to what we require, but the earliest point at which outcomes are measured is 6 months after the surgery. Another appears to be exactly what we require as it provides data about mean walking ability from 1 week to 1 year after the surgery.3 Throughout the rest of this chapter, we will refer to this study as the ‘knee arthroplasty study’.
Clinical scenario (continued): Structured abstract of our chosen article (the ‘knee arthroplasty study’)
Citation: Kennedy D, Stratford P, Riddle D et al. Assessing recovery and establishing prognosis following total knee arthroplasty. Phys Ther 2008; 88:22–32.
Question: In clients undergoing a primary knee replacement for osteoarthritis, what is the pattern of improvement in lower limb function and walking ability from 1 week to 1 year after surgery?
Design: Inception cohort followed prospectively for 1 year.
Setting: Tertiary care orthopaedic facility in Toronto, Canada.
Participants: Eighty-four clients with osteoarthritis (mean age 66 years, 52% female) undergoing primary total knee arthroplasty. Participants needed to be able to communicate in written and spoken English. Exclusion criteria were any neurological, cardiac or psychiatric disorders or other medical conditions that would substantially compromise physical function.
Prognostic factors: Gender, pre-operative lower limb function and pre-operative 6-Minute Walk Test distance.
Outcomes: Lower Extremity Functional Scale—a self-reported, 20-item scale of lower extremity function that includes activity limitation and participation restriction concepts and is scored from 0 (lowest function) to 80 (highest function). Six-Minute Walk Test—a submaximal exercise test in which participants walk the greatest distance they can in 6 minutes on flat ground with standard encouragement.
Main results: In general, there was a deterioration in both scores at the immediate postoperative measurement. From this point, there was rapid improvement in both scores initially, with progressively slower gains in improvement with increasing time postsurgery. Gender and pre-operative function were prognostic factors for each outcome. For the Lower Extremity Functional Scale, assuming the average preoperative function of the cohort, females had scores of 18 at 1 week, 38 at 6 weeks, and 53 at 6 months. Males had scores of 25 at 1 week, 43 at 6 weeks, and 60 at 6 months. For the 6-Minute Walk Test, assuming the average preoperative function of the cohort, females achieved 200m at 1 week, 330m at 6 weeks, and 470m at 6 months. Males achieved 250m at 1 week, 440m at 6 weeks, and 580m at 6 months. All the 6-month values were maintained to 1 year.
Conclusions: For both outcomes, the rate of improvement was greatest immediately following surgery. Roughly half of the postoperative improvement was observed in the first 6 weeks. This brought clients to an adequate level of function for typical daily activities. Almost all of the remaining improvement had occurred by 6 months, with both outcomes then being maintained until the end of the year.
Is this evidence likely to be biased?
We will use questions drawn from the Critical Appraisal Skills Program (CASP) and associated checklists for appraising a cohort study to explain how to assess the likelihood of bias in a prognostic study. Note, however, that the checklist for cohort studies is not only intended for use with longitudinal single-group studies, but also with other study designs such as case-control studies. Therefore, not all the questions that are raised in the checklist will be explained in this chapter. The key questions to ask when appraising the validity of a prognostic study are summarised in Box 8.1. The checklist begins with two simple screening criteria that, if not met, indicate that the article is unlikely to be helpful and that further assessment of potential bias is probably unwarranted.
Did the study address a clearly focussed issue?
The first criterion on the checklist is whether the study addressed a clearly focussed issue. For prognostic evidence, the article should clearly define the population, potential prognostic factors and the outcomes considered.
Clinical scenario (continued): Did the study address a clearly focussed issue?
The knee arthroplasty study meets this criterion, as shown in the question it seeks to answer: In clients undergoing a primary knee replacement for osteoarthritis, what is the pattern of improvement in lower limb function and walking ability from 1 week to 1 year after surgery?
Appropriate study type
The second criterion is that the method used was appropriate to answer the question posed by the authors. In Chapter 2, we saw that longitudinal studies, particularly prospective cohort studies, provide the best evidence about prognosis. Even better than that is a systematic review of prospective cohort studies. However, currently there are so few systematic reviews of prognostic studies in this area that it is probably not realistic for you to expect to find one.
Although prospective cohort studies are typically the study type that you should use to answer prognostic questions, you should be aware that prognostic information can also be generated by other study designs. For example, if you are interested in the natural history of a condition, then the outcomes of an untreated control group in a randomised controlled trial can provide this information. Conversely, case-control studies or case series, where all cases receive a particular treatment, give prognostic information about a treated cohort.
Clinical scenario (continued): Appropriate study type
The researchers in the knee arthroplasty study wanted to find prognostic data and they used the ideal study design for this: a prospective, longitudinal study of an inception cohort of clients undergoing primary total knee arthroplasty. Therefore, we can move on to the more detailed criteria on the CASP checklist.
Representative and well-defined sample of participants
The next criterion on the checklist is whether the cohort was recruited in a way that ensured it was representative of the larger population of interest. This criterion is important as a study’s estimate of prognosis will be biased if the study’s sample is systematically different from (and therefore not representative of) the larger population of interest. It is important that a study clearly defines its inclusion and exclusion criteria as this can help in recruiting a representative sample. Clearly defined criteria help make it clear to everyone (researchers, participants, you) just what the target population of the study was. A representative sample is also more likely to be obtained if the study recruits all of the eligible clients who presented at the recruitment site into the study. When appraising a study, look for a statement in the article that describes either recruiting ‘all clients’ or recruiting ‘consecutive cases’. Recruiting all eligible clients prevents bias in the data that could arise if some eligible clients avoided recruitment and these clients differ in some systematic way from those who were recruited. The greater the proportion of eligible clients that are recruited into the study, the more representative of the target population the sample is likely to be.
Clinical scenario (continued): Representative and well-defined sample
In the knee arthroplasty study, the researchers were unable to achieve consecutive recruitment because of the outbreak of severe acute respiratory syndrome (SARS) in Toronto during their data collection period. However, we are not too concerned about this as it is unlikely to have caused recruitment of an unrepresentative sample. Additionally, the inclusion and exclusion criteria of the study are clearly defined. We conclude that the cohort is likely to be representative of the target population.