Chronicle of a Death Foretold

Introduction

Gabriel García Márquez opens Chronicle of a Death Foretold with a brutal finality: a man is murdered in the doorway of his home. Hundreds of pages later, the reader has some insight into why the death happened, but the event itself is never in doubt.

Like the famous novel, the mortality follow-back study is a subset of case design, constructed in reverse time: collect a set of patients who have some outcome—in this case, death—under some defined set of circumstances or with some defined condition, and look back to understand the days, weeks, or years leading up to that event. This study design is abundant in the cancer literature to understand end-of-life care patterns, costs, and experiences of death (1,2). More recently this design has been applied to the ESKD population. This is a population where mortality rates exceed those of the major cancer subtypes (3), so patient death is a common event, and one of great interest to clinicians and epidemiologists alike.

These studies appear to answer important questions about the dying process. They have been published by prominent journals and are then reflected in prominent news media (4) to highlight the often agonizing deaths of patients with ESKD; these are typically contrasted with the gentler options of conservative management. However, the conclusions drawn from them are only valid within the framework of case-series studies, with the “case” definition here being death with a certain set of diagnoses, and comparison often made to another group of decedents with different diagnoses.

For many years, concerns have been raised about the biases and applicability of such studies. Bach et al. published an elegant rebuke to the use of this study design in the cancer literature (5). Here, I adapt their observations to the role of this study design in the kidney disease literature. I review Bach et al.’s critical distinction between dying and dead, and argue that because ESKD is a diagnosis with highly variable survival, the follow-back design is fraught with bias.

First, consider studies focused on the experiences of a group of decedents in the months or years before death. Study populations are selected using an outcome-based definition, as in a case-series design: dead after invasive procedures with a diagnosis of ESKD (6), or dead in the year after amputation with a diagnosis of ESKD (7). In the latter study, which I will refer back to as an example, decedents with ESKD and an amputation in the year before death are compared with ESKD decedents without amputation, and to decedents without ESKD who had or did not have an amputation. These differ from traditional cohort studies, where patients are grouped by their diagnosis or exposure (a group of ESKD patients undergoing an amputation versus a group not undergoing amputation, for this example) and then followed for a period of time thereafter until the event of interest (death) occurs or the study ends. The case series is a study of dead patients; the cohort study is a study of potentially dying patients.

The study of dead patients provides the opportunity to obtain detailed information about the events before death in these defined populations, which has its own value. For example, we learn that patients undergoing amputation in the last year of life have more hospital days and are more likely to die in the hospital. We learn that, compared with non-ESKD decedents, amputation was more common in the last year among those with ESKD.

Our general impulse on reading such a study is to apply it in some way to improve the lives of our patients with similar conditions; for example, we might be persuaded that patients with ESKD and peripheral arterial disease have palliative care needs that we are not addressing, or even that pursuit of aggressive intervention such as amputation is futile. This certainly seems logical in the context of a study of dead patients. But would we draw similar conclusions from a cohort study as we do from this study? In this case, is potentially dying the same as dead?

Because medical care occurs along a diagnosis-to-outcome time course, case studies defined by the outcome of death are clinically helpful to those with a given diagnosis (e.g., ESKD and an indication for amputation) only if certain facts hold about the relationship between diagnosis and death. First, it must be true that the diagnosis is closely tied in time to the outcome. If the diagnosis might have occurred at prior time points, then some people with this illness will have been missed. Second, the diagnosis must be closely tied in certainty to the outcome. People with this diagnosis must progress both quickly and inevitably toward death, otherwise the basic presumption that studying those who died of the disease represent those diagnosed with it is lost.

The left side of the figure (panels A and C) shows a scenario in which the assumptions above are met, and studying decedents with the condition of interest in the year before death creates an identical study population compared with studying those diagnosed with the condition of interest. Surprisingly, it is challenging to identify a situation that meets these assumptions but is not obviously futile care. A variety of scenarios such as mechanical ventilation in stem-cell transplant recipients (8) have been cited, but survival rates subsequently improved and these situations became less applicable (9).

The right side of the figure (panels B and D) illustrates, instead, the scenario in which these assumptions are not met and specifically the conditions used to define the decedents in the study of patients with ESKD with lower extremity amputation (7). Patients with ESKD undergoing amputation do have high mortality rates, recently reported in cohort analysis at about 44% over a year after amputation (10); however, amputation is clearly neither a time-stamp of an exact survival window (because peripheral arterial disease is not inexorably fatal over a short course) nor a certain precursor to death (because we have good evidence that 56% of patients survive more than a year after amputation). The very fact of undergoing amputation suggests that these patients were considered not to be moribund. In this situation, studying those who have died (dark gray bars) does not help differentiate provision of care for those in this situation, who comprise the majority of those undergoing amputation, as illustrated by the light bars, captured in the cohort study simulation in the top right but lost to the case study in the bottom right. The cohort study here is a much better reflection of the risks and benefits of the intervention and provides a built-in comparison group that is more clinically relevant than the case studies can provide. The presentation of data in the case study, reflecting only those who died, both inflates the impression of futility for those considering amputation and ignores the benefit of amputation and aggressive medical care for those who survive. Similarly, finding that younger patients with ESKD undergo more invasive procedures and endure more intensive care unit (ICU) time in the months before death (6) elides acknowledgment of the likelihood of a successful outcome.

When spelled out in this frame, the biased inferences that result from examining case studies of decedents seem clearer. Those in favor of the study design assert there are few good ways to study and improve the dying process (1), and in limited situations I agree. These studies can help explicate the family experience of end-of-life care. By combining knowledge of actual mortality rates from the condition at hand with knowledge of the potential course of the dying process, we can become better guides for patients with ESKD (11). But it is difficult to accept at first glance how challenging the interpretive task is for this study design and how hard these data are to apply to the practice of medicine.

As a dialysis physician, I am not naive to the idea that my patients are at high risk of death; indeed, many deaths have occurred under my care. But suppose that I were responsible for deaths alone, with my aim being to avoid hospital-bound deaths. I might group together those who died in the ICU and scrutinize what I did wrong to allow death to occur in this setting. I might also group together those who died at home or with hospice. How did I fulfill their wishes for a peaceful death?

The reality, however, is that these outcomes do not adequately represent the lives that preceded them. Among those who died in the ICU, there were those who ought to have survived, patients for whom our focus was quite appropriately on attempts to save their lives. These patients would be miscategorized by a case-series approach as having unduly suffered during a last year of life, when in fact their ICU stay was our best attempt at appropriate care. Among those who died with hospice, some suffered great pain in the years before the “peaceful” death that is on record.

Most critically, lives saved are not accounted for by this reckoning. The patient who survived the ICU and recovered from bypass surgery was not an ICU death, so does not meet the “case” definition of patients with ESKD dying in the ICU after a procedure. The man with the necrotic leg who survived the amputation and is now living a fulfilling life despite his ESKD is not represented. The patient who started a time-limited trial of dialysis in the ICU that has now lasted several years is not included. None of these successes will be reflected in a study of decedents—at least not yet—and so none can counterbalance the negative impressions left of intensive medical care for patients with ESKD.

The follow-back study design, then, gives answers that are technically accurate with respect to the decedent population, and can illuminate the experience of dying with ESKD, but they provide no meaningful guidance for clinical care for all but the most predictably terminal scenarios. Survival rates in most scenarios in ESKD are well above a futility threshold and are unpredictable on an individual basis, and so the data available from follow-back mortality studies provide limited insight into decisions about clinical care.

Because the deaths we chronicle in ESKD are rarely foretold, I suggest we follow our oncologic colleagues and better delineate the role and limitations of case-based studies of decedents. The design of these studies should be clearly defined, and survival rates from the conditions under study should be part of the discussion. For most scenarios, other study designs can better characterize our living patients and their wishes and preferences.

Figure 1.

Selection biases possible in look-back studies of mortality versus traditional cohort studies. The left side of the figure (A and C) shows a scenario where all patients with a given condition die within a year of identification. In this case, cohort and look-back studies capture the same individuals (dark gray). The right side of the figure shows a scenario where instead 30% of those with a given condition die within a year. Here, all patients are captured by the cohort design (B), whereas only those in dark gray are captured by the look-back design (D).

Disclosures

Dr. Rifkin is supported by a Veterans’ Affairs Merit Award.

Funding

None.