Evidence-Based Prognosis

As the Case Study demonstrates, discussion of survival after stroke can be challenging. Although prognosis was often part of medical training in the early 1900s, the ever-burgeoning body of clinical medical knowledge has slowly nudged this from most medical programs. Major advances in the past 50 years have focused more on extending life, with quality/quantity of that remaining life often seen as lesser priorities.1,2

In the modern context, prognosis pays a central role in oncology, and cancer survival papers are numerous. For each type of cancer there are substantial resources documenting the 1- and 5-year survival rates, often stratified by a person’s age, sex, race, tumor stage, grade, histology, and treatment. As a result, physicians routinely quote survival and relapse rates to their patients with cancer. There are also numerous mortality risk indices for general use by clinicians in medical decision making for the geriatric population.3 Until recently nothing remotely similar existed for stroke, although it is the fifth leading cause of death in the US, second leading cause of death in the world, and the leading cause of disability.

Survival Prognosis After Stroke

A recent article4 on long-term survival after stroke provides life expectancies by age, sex, and severity of disability as measured by the modified Rankin Scale (mRS), the most widely used measure of long-term disability after stroke. The life-expectancy results in the article were based on a systematic review of the stroke literature including 11 key studies comprising more than 35,000 individuals followed for at least 1 year. Prognosis was made from at least 3 months poststroke, following the acute period of increased mortality risk and after most structured rehabilitation programs. All types of strokes were considered, along with possible comorbid factors.

As indicated in the study, it is a matter of empiric observation that age and severity have the largest effects on long-term survival. For example, male individuals had an overall hazard ratio (HR) of approximately 1.2 compared with women, indicating 20% higher mortality risk; the risk approximatly doubles (HR=2.0) with each decade of age; and the largest cohort study5 evaluating survival reported that the effects of diabetes, smoking, and atrial fibrillation each had an HR of approximately 1.4. By contrast, the very same studies show that those with mRS grade 4 have nearly 3 times the risk compared with mRS grade 0, and those with mRS 5 have 5 times the risk. If factors other than age, sex, and severity are ultimately deemed relevant, these too can be considered in the analysis.

We believe the complete set of results from the study, including the underlying survival probabilities, together with some practical instruction regarding prognosis, can effectively be used by clinicians to provide evidence-based guidance on long-term survival after stroke. To facilitate this, we provide a web-based survival tool for rapid and accurate calculations at http://www.LifeExpectancy.org/asp/stroke/.

Survival Prognoses In General

Life expectancy is a precise technical term referring to the average (arithmetic mean) survival time in a group of similar persons. Life expectancy is generally derived using a life table; see the Box for a refresher on this scientific tool.

While colloquially the term life expectancy may be used to express a summary view about a patient’s prognosis, it is not meant to be a prediction about a given person’s actual survival time. For example, the general population life expectancy of males age 70 is 14 years, but death much sooner or instead after twice as long is entirely possible. Similar patients can live very different lifespans. For example, 1 identical twin may outlive the other by 30 years. It is a truism that an individual patient’s survival time cannot be predicted with any reasonable degree of precision. It is not merely difficult; it is impossible (Box Figure 3).

To give an opinion on an individual’s likely prognosis requires 1) sufficient information to identify factors, features, or characteristics of the individual known to be related to survival and 2) identified groups (cohorts) of similarly situated persons’ actual survival experience. Prognosis requires matching the person in question to the survival of similar persons. Medical doctors routinely perform this sort of comparison when making diagnoses, identifying treatment options, and deciding upon surgical techniques. In the simplest life-expectancy context, we merely identify the appropriate patient groups for comparison, then report on their empiric survival.

In our experience, most medical doctors feel confident in providing survival prognoses only when a person is at the extremes of the survival spectrum—when the prognosis is poor, on the one hand, or very good on the other. When prognosis is poor and medical intervention is frequent and often invasive with long hospital stays, patients and family may already essentially know the prognosis, and it need not be quantified with any precision. In the latter case, the clinician may feel that the condition does not affect the person’s mortality risk and therefore does not affect life expectancy. In that case, mortality risk and life expectancy may be said to be normal, and general population figures can be cited, as appropriate, to support this statement. In the remaining intermediate cases, which are quite common in practice, the clinician may not have any known guideposts and, thus, may be reluctant to offer an opinion. For example, clinicians know that smoking reduces life expectancy, but few would know by how much, or how it depends on the number of pack-years. More difficult still are cases with multiple complex risk factors, such as heart disease, diabetes, and obesity.

A busy practicing clinician would not be expected to have or to develop the scientific expertise to make accurate calculations of life expectancy in every case, even in their own medical discipline. Furthermore, survival in some conditions has changed much in the past 20 years, making knowledge gained in medical school less applicable. Such calculations are squarely in the realm of medical directors, life actuaries, underwriters, and medical researchers. Yet some medical doctors (especially oncologists) do routinely use specific literature from their given field of expertise to provide at least approximate survival figures. We believe that the same can be done for stroke cases.

Survival Prognoses After Stroke

The Table reproduces the summary life expectancy figures from the aforementioned article on long-term survival after stroke.4 For a man age 70, such as the person described in the opening Case Study, whose disabilities are consistent with Rankin grade 4, the stated life expectancy is 6 additional years. That is, amongst a group of similarly situated men, the arithmetic mean, or average, survival time is 6 years.

The fact that the life expectancy for this group is less than the corresponding general population figure of 14 years indicates that Mr. P’s life expectancy is not normal. But this single life expectancy figure does not specify with certainty exactly how long he will live.

In the present context, we are concerned with giving an accurate and helpful description of survival that is tailored to the case at hand. Broad statements such as those offered in the Case Study, although correct, are not entirely helpful. More precise statements are usually needed, although presenting the average for the group may not be sufficient. Indeed a single figure cannot provide a complete description of a person’s prognosis for survival. What is required are all the details concerning survival. These are embodied in the probabilities of surviving for 1, 2, or more years, which are, in fact, given in the commonly known survival curves, frequently presented in the medical literature (Box).

Calculating Survival Curves After Stroke

The Figure shows survival curves for 70 year-old males from an online calculator (http://www.LifeExpectancy.org/asp/stroke/) based on the same published research shown in the Table. The calculator produces survival curves based on any combination of age, sex, and disability. As noted, age and severity of disability are the 2 factors with the largest effect on survival after stroke. Sex is also included because doing so is standard in most survival analysis, and indeed being female is a modestly positive factor amongst persons with mild or no disabilities following stroke.

<p>Figure. Survival Curves By Rankin Grade: 70 Year-Old Men</p>

Click to view larger

Figure. Survival Curves By Rankin Grade: 70 Year-Old Men

In the Figure, there are 7 curves, one for each of Rankin grades 0 to 5 and the general population (GP). Notice that each curve starts at time 0 (age 70) with 100% of persons alive and then decreases with the passage of time. Also notice that the curves for more severe disability are uniformly lower than those for less severe disability, reflecting the lower survival at each age and time.

The red curve (second from the bottom) is for Rankin grade 4, which applies to Mr. P and provides exact survival values, detailed in the table below the graph. For example, 79% of people survive 2 years, 61% survive 3 years, …, 5% survive 16 years, and only 1% survive 20 years.

In the highlighted row, for Rankin grade 4, observe that 100%–79%=21% will live 0 to 2 more years, 18% will live 2 to 4 years, 16% will live 4 to 6 years, 13% will live 6 to 8 years, 10% will live 8 to 10 years, 8% will live 10 to 12 years, 6% will live 12 to 14 years, and 8% will live 12 or more years. The sum of these 8 probabilities is 100%, as appropriate.

Notice that the 2-year time interval that captures the most likely survival time is 0 to 2 years; yet only 18% (less than one-fifth) of the group falls into this bin. To capture more of the survival times, years 0 to 4 could be used, but this would contain only 39% of the observed values.

The median survival time—also known as the more likely than not survival time—is the time after which exactly 50% of the initial group are still alive. By interpolation from the data in the Figure, this is 5.4 years. That is, 50% of similar patients will survive at least 5.4 additional years, and the other 50% will have passed away prior to this time point.

Practical Application

What then should be said to Mr. P and his family? Of course, such discussions are both art and science. Much has been written about the former; we concentrate here on the latter: what can be said, rather than on how to do so.

1. Identify the appropriate Rankin Group. For Mr. P it is grade 4, because he cannot walk but retains some ability in self care.

2. Observe that median survival time for males age 70 in this group is 5.4 years. That is, Mr. P has a 50% chance to survive roughly 5 more years.

3. Note that amongst men like him, the average survival time (ie, the life expectancy) is 6 years.

4. Explain that Mr. P has an 8% chance of surviving at least 14 years, which is the average survival time for men his age in the general population. Because he does not have a normal life expectancy, it is not surprising that he has a reduced chance of living what would be the normal amount of time.

5. Convey what is known about the entire survival distribution for men like him— all of the survival possibilities and probabilities—if that is of interest. For example, the family may have a strong desire to care for the patient at home, although understanding that the burden of doing do may not be sustainable for more than 2 years. Their understanding that the patient is quite likely (79%) to live more than 2 years may well affect their decision-making.

It may be, of course, that a person has additional risk factors that make prognosis worse than average amongst the Rankin grade 4 group, or perhaps conversely, no other risk factors, making prognosis more favorable for the individual. In such cases the calculated figures would be either somewhat pessimistic or optimistic. Nonetheless, they provide a rational starting point from which further scientific adjustments could be made.

Limitations and Conclusions

As can be seen, there are limitations inherent in the science. Most obviously, no specific survival time, even if rounded to the nearest integer, or even any narrow range of such figures, will occur in 100% of cases. As noted, it is impossible to predict accurately an individual person’s exact survival time except in the most unusual of circumstances. Also, in the case of stroke and other neurologic injuries it may be prudent to postpone discussions of long-term prognosis until after functional recovery has plateaued, as a proportion of persons with even exceptionally severe acute strokes may actually return to baseline.

Further, there are limitations in the practitioners. As documented by Nicholas Christakis, MD, PhD, in Death Foretold, predictions of survival by medical doctors are known to be systematically optimistic, and more so the better the doctor knows the patient.6 By contrast, as explored more than 50 years ago by Paul Meehl, evidence-based predictions have been documented to be generally superior to clinical ones.7-9

At the same time, a clinician can often identify factors, features, or characteristics of a given person that make them better or worse than average amongst an otherwise similar group, suggesting that the nominal figures may require some adjustment. In economic circles this is known as tailoring. In such cases the empirical evidence regarding the effect of known factors is considered before accounting for other factors.

The solution for the practitioner, then, is to rely on the scientific evidence as possible, despite its inherent limitations, in order to convey to the patient what is known about the survival of similarly situated persons. The figures and tables here, together with the online calculator, can provide clinicians with the best currently available evidence on the prognosis of their stroke patients.

1. Brown GC. Living too long: the current focus of medical research on increasing the quantity, rather than the quality, of life is damaging our health and harming the economy. EMBO Rep. 2015;16:137-141.

2. The Lancet Respiratory Medicine. Prolonging life at all costs: quantity versus quality. Lancet Respir Med. 2016;4(3):165.

3. Lee S, Smith A, Widera E, Yourman L, Schonberg M, Ahalt C. What is ePrognosis? ePrognosis. Available at: https://eprognosis.ucsf.edu/about.php. Accessed January 9, 2020.

4. Shavelle RM, Brooks JC, Strauss DJ, Turner-Stokes L. Life expectancy after stroke based on age, sex, and Rankin grade of disability: a synthesis. J Stroke Cerebrovasc Dis. 2019 Dec;28(12):104450.

5. Eriksson M, Norrving B, Terént A, Stegmayr B. Functional outcome 3 months after stroke predicts long-term survival. Cerebrovasc Diseas. 2008;25:423-429.

6. Christakis NA. Death foretold: Prophecy and Prognosis in Medical Care. 1999; Chicago: University of Chicago Press.

7. Meehl PE. Causes and effects of my disturbing little book. J Pers Assess. 1986;50:370-375.

8. Meehl PE. Clinical Versus Statistical Prediction: a theoretical analysis and a review of the evidence. 1954; Minneapolis, MN: University of Minnesota Press. Reprinted by Echo Point Books & Media, 2013.

9. Dawes RM, Faust D, Meehl PE (1989). Clinical versus actuarial judgment. Science. 1989;243:1668-1674.

RS, JB, DS, and AT report no disclosures.