Supportive Care: Time to Change Our Prognostic Tools and Their Use in CKD

Usefulness of Prognostic Tools

Prognostic tools provide an estimate of how many people with similar risk factors will experience a given event (for example, death), but they cannot identify who will experience this given event.

The Renal Physician Association (RPA) and the American Society of Nephrology in their clinical practice guideline on shared decision making in the appropriate initiation of and withdrawal from dialysis recommend that “[a]ll patients requiring dialysis should have their chances for survival estimated, with the realization that the ability to predict survival in the individual patient is difficult and imprecise” (9). This patient-specific estimate of prognosis aims to facilitate informed consent in the process of shared decision making and support the discussion of supportive care goals (9). Prognosis of the course of an illness is among the main responsibilities of medical doctors (10). Studies have shown consistently that patients want to have the nephrologist discuss their prognosis but that this is rarely done (11–13). Prognostic tools could help health professionals identify high-risk patients and facilitate discussions regarding adapted care (for example, potential imminent need for supportive care), ethics advice, or treatment adjustment. Overoptimistic estimations of prognosis may lead to inappropriately disease–directed treatment (14,15). Validated prognostic scores may also facilitate nephrologist recommendation of dialysis services to those most likely to benefit and proposal of alternative care pathways that might better respect patients’ values and goals to those unlikely to benefit (16).

Prognostic tools could also help in shared decision making with patients and their loved ones, who have the right to be informed and part of the decision-making process in any decision concerning their treatment. Decision support interventions help people consider the choices that they face, they describe where and why choice exists, and they provide information about options. Prognostic tools would be especially useful when conservative management is discussed in the various following situations: in advanced CKD care with the incorporation of palliative care principles and without the initiation of RRT (17), in palliative dialysis (3), or in dialysis withdrawal (18). Evidence-based information about outcomes improves people’s knowledge regarding options, reduces their decisional conflict related to feeling uninformed and unclear about their personal values, and improves accurate risk perceptions when probabilities are included (19). Especially in older adults, information about prognosis may help patients plan for their remaining lifetime.

Current Tools Available

In the past, prognosis was primarily guided by clinical acumen. Later, ancillary information, such as laboratory data and imaging studies, was added. Over the years, the sheer amount of information related to the patient and his/her likely trajectory became vast, and systematic approaches to integration of data from multiple sources became a necessity. Although this patient population has frequent encounters with medical staff, which should, in theory, assist in prognostication (by increasing the clinician estimate of prognostication), the process is hindered by the realities of patient frailty and comorbidities. In the process of shared decision making, gut instinct is usually complemented by an analytic approach, where deliberation of multiple factors takes center stage (incorporating laboratory and demographic data other than the clinician’s estimate). Prognostic scores are meant to support this analytic process.

In the past 10 years, numerous scores of mortality on dialysis have been developed on the basis of various combinations of comorbidities and laboratory data. Although early mortality scores are important to consideration of end of life support or requirement for hospice (4), only a few of them focused on short-term survival (≤6 months) (20–24), some of which included only elderly patients on dialysis (21–23). These prognostic scores are summarized in Table 1. Other factors, like self–rated health question (25), perceived treatment control (26), patients’ illness perception (27), Karnofsky performance score (28), or frailty (29,30), are related to survival, but their prognostic value for short-term mortality (≤6 months) has not yet been evaluated. Other scores were developed to predict progression to kidney failure in patients with CKD stages 3–5, but they usually predict a long-term risk (≥1 year) and may not be directly useful in the context of when to incorporate kidney supportive care (31–33).

“The use of objective measurable data in statistical predictions is more accurate than the subjective impressions or intuitions or clinical judgment of trained professionals” (Daniel Kahneman, Nobel Prize winner). However, clinical intuition may add value when two basic conditions are met: an environment sufficiently regular to be predictable and an opportunity to learn these regularities through prolonged practice (34). An example is given by the integration of the surprise question in a score that also includes objective measures (24). The discrimination value of this score developed by Cohen et al. (24) is higher than other scores, although predictive power varies with clinical discipline, seniority, and clinical setting (35).

Use of Prognostic Tools in Daily Practice

Although for individual patients, current prognostic tools are not sensitive or specific enough to tell a patient how long they will live, they are informative at identifying high-risk populations. This group of patients can then be targeted for specific interventions that can potentially benefit them. The prognostic tools can also be used alongside decision aids to help in the shared decision–making process for individual patients (36).

Multivariate predictive scores provide death probabilities and probability cutoffs that can be chosen as relevant for a given clinical context (e.g., predictions for considering a more palliative approach to care versus predictions that would suggest that patients may have a higher likelihood of benefitting from various life–prolonging interventions). They can be automated, and their administration does not have to consume additional physician time (except the surprise question). Currently, these tools are available for free online (https://qxmd.com/calculate/3-month-mortality-in-incident-elderly-esrd-patients) and also, on mobile applications (calculated by QxMD). They also can be used along with the RPA Shared Decision Making toolkit (https://itunes.apple.com/us/app/rpa-sdm-toolkit/id843971920?mt=8). Figure 1 illustrates how a prognostic tool can be incorporated into a decision process for kidney supportive care.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Role of the prognostic tool in kidney supportive care decision. *Search for geriatric syndromes (such as dementia, functional disability, malnutrition or depression) and assessment of patients' psychosocial and spiritual situations.

There is early experience with the use of prognostic tools in kidney programs, especially in the United Kingdom and Canada, where these tools are being used primarily to help identify high-risk patients for additional supportive care. However, a recent national survey of Canadian nephrologists showed that >80% of the respondents were not satisfied with their current ability to predict clinical trajectories and that this ability was perceived to be very important for decision making. Furthermore, there was strong support for the further development and uptake of validated prognostic tools to enhance appropriate care that is aligned with patients’ priorities and illness trajectory (37).

Limitations or Deficiencies of the Current Tools

The process of prognostication is difficult and complex because of patients’ multiple comorbidities and overlap of different prognoses. The global health of patients with CKD includes many domains, like comorbidities as well as functional impairment, geriatric syndromes, and social environment. All of these domains have to be taken into account because of their potential effect on outcome and quality of life. Therefore, new prognostic scores that integrate these domains need to be developed.

Recent work has identified that enhanced communication among patients and providers, especially around dialysis modality options, is a top Canadian patient priority (38). This will require a better understanding of illness trajectories, focusing on not just survival but also, other outcomes equally relevant to patients, such as the effect of treatments on HRQOL and physical function, hospital-free survival, place of death, and capabilities (8). In fact, current predictive tools are not necessarily patient focused and do not necessarily address their concerns or their family concerns. Clinicians should be trained to better communicate prognosis in particular by tailoring the conversation to the individual patient’s preferences for information and participation in the informed consent process (5,6). For patients with complex cases, a specialist palliative care consultation can be very valuable to organize and support this prognosis communication. The access to outpatient specialist palliative care, limited in many countries, should be improved.

To help the integration of those tools into clinical practice, it is important to clarify when those scores should be used, which score should be used in which context, how these estimates should be communicated to the patients, and how they can be integrated in a shared decision–making process. For example, we recently proposed a risk stratification algorithm to screen for evaluation, evaluate, and decide on an appropriate strategy of care for elderly patients with ESRD according to their level of risk of early death (21). Such an approach illustrates the usefulness of prognostic tools to decide the right timing of starting kidney supportive care.

Sentinel Events and Biomarkers in Patients on Dialysis

Recent research has indicated that death in patients on dialysis is preceded by characteristic events, such as hospitalizations (39) or a decline in HRQOL scores (S. Johnstone, et al., unpublished data). The dynamics of laboratory and clinical changes seem to accelerate in the terminal 6 months before death as shown by analysis of their rates of change (Table 2). Decrease in albumin level (40,41), BP (41), and body temperature (42) on the one hand and increase in glucose (Calice-Silva, et al., unpublished data), C–reactive protein levels (41), and neutrophil-to-lymphocyte ratio (43) on the other hand are associated with an increased risk of death. In addition, a reduced nutritional score (44) and worsened nutritional competence as indicated by lower body weight (40,45), interdialytic weight gain (41,43), phosphorus (43), lean body mass, and fat mass on one hand and increased fluid overload (A. Guinsburg, et al., unpublished data) on the other hand are also associated with an increased risk of death. Increased inflammation and poor nutrition translate into distinct alterations in body composition, with a loss of lean body mass and fat mass. Although the pathologic pathways underlying these sentinel events remain to be determined, it seems that a surge of inflammation may provide a reasonable unifying concept (41).

Lessons from Other Populations

Numerous scores have been developed in populations other than patients with CKD. These scores typically use a combination of symptoms, syndromes, or functional measures that could be relevant for patients with CKD. Future research needs to focus on creating more robust models that look at these additional variables and evaluate their performance in patients with CKD.

In the general population, the Patient–Reported Outcome Mortality Prediction Tool, which assesses sociodemographic characteristics, health and lifestyle behaviors, comorbidities, activities of daily living, and HRQOL, is a promising tool to predict 6-month mortality in elderly patients (46). In patients with cancer, low performance status or its deterioration (47,48) and the Clinical Prediction of Survival (15,47,49,50) are associated with short-term mortality. Numerous scores with good discrimination performance have been developed in the context of palliative care. They include a combination of the following factors: Karnofsky performance scale, ambulation, activity, evidence of disease, self-care, oral intake, edema, dyspnea at rest, delirium, level of consciousness, clinical prediction of survival, total white blood cell count, and lymphocyte percentage (47,48,51).

Patients’ Priorities

Patient-centered care involves shared decision making and disease management between a patient and his/her clinician. A shared decision–making process can ultimately encourage adherence to treatment and improve patient-centered outcomes (52). Delivering patient-centered care requires an understanding of outcomes that are important and relevant to patients. However, the CKD research agenda has been driven traditionally by investigators, resulting in studies that may not optimally inform patient care or may not show priorities for those who are affected by the condition (53,54). The importance of engaging the broader stakeholder community, including patients and their caregivers, in determining research priorities and actively participating in the research has been recognized. Risk of hospitalization, HRQOL, cost, and intercurrent events may be other priorities to the patients in addition to survival (8). A recent systematic review (55) of research priority setting in kidney disease identified 16 studies that elicited patient, caregiver, or health care provider priorities for research in kidney disease; of those, only four (25%) explicitly involved patients, and only one included patients with early stages of CKD not yet on dialysis. For example, a recent study has shown that patients tend to prioritize outcomes that are more relevant to their daily living and wellbeing, like ability to work, whereas caregivers ranked mortality, anxiety, and depression higher than patients did (56).

Despite efforts to increase patient involvement in research for more than a decade, there has been little progress (57), and recent studies continue to identify a lack of patient engagement in the research priority setting (58). More recently, steps have been taken to ensure patient engagement in research and the research priority setting, with funding agencies in several countries now mandating patient engagement as a requirement for funding consideration: the Patient–Centered Outcomes Research Institute (PCORI; http://www.pcori.org/research-we-support/pcor/) in the United States, the INVOLVE Project (http://www.invo.org.uk/about-involve/) in the United Kingdom, and the Strategy for Patient Oriented Research (http://www.cihr-irsc.gc.ca/e/41204.html) in Canada. For example, the PCORI is supporting research of a peer–led mentoring program and home-based care, whereas the INVOLVE Project supported research on a self-management intervention and the effect of social media.

Predictors of Quality of Life, Hospitalization, and Functional Decline

Frailty affects both physical and mental quality of life in predialysis patients with CKD (59,60). Frailty measured with five components (shrinking, weakness, exhaustion, low activity, and slowed walking speed) is an independent predictor of number of hospitalizations (61). Gait speed is a predictor of hospitalization and functional decline (difficulty with activities of daily living and estimated change in Short Form [36] Health Survey Physical Function score) at 12 months in prevalent patients on hemodialysis (62). Self-reported appetite is significantly associated with several measures of inflammatory and nutritional status, and poor appetite is associated to lower subjective HRQOL and increased rates of hospitalization and mortality during a 12-month follow-up period (63). The contribution of those predictors in prognosis score should now be evaluated.

In other populations, weight loss, fatigue, and markers of systemic inflammation are associated with adverse HRQOL, reduced functional abilities, more symptoms, and shorter survival (64). In particular, fatigue is a common symptom associated with the presence of inflammation, poor quality of sleep, depression, anxiety, and poor HRQOL (65). Physical frailty indicators can predict disability in activities of daily living in community–dwelling elderly people (66). The screening tool Identification of Seniors at Risk–Primary Care has moderate predictive value in identifying persons at increased risk of functional decline in community-dwelling persons aged ≥70 years old (69) or elderly patients presenting to an emergency department (68,69).

Ethical Aspects

To ensure that patients are not unfairly discriminated against, prognostic tools should not be used (1) to deny a treatment, (2) across populations without consideration of context, (3) beyond the boundaries of the derivation or validation cohort without recognition of limitations, (4) without appreciation of the uncertainty of the estimates of prognosis, (5) to irreversibly classify or categorize a patient without appreciation of the possibility of a change in status, and finally, (6) in isolation without considerations of clinical context.

Generalizability

No matter how well calibrated and discriminating (Figures 2 and 3) a prognostic tool may be in development, a tool that can only predict outcomes in the sample in which it was developed is useless. For a system to be generalizable, the accuracy of the tool must be both reproducible (accurate in an identical population) and transportable (70). Transportability is shown if the system is accurate in patients drawn from a different but related population or data collected by using methods that differ from those used in development. Historical and geographic accuracy is maintained when the system is tested on data from different calendar times or different locations. Methodologic accuracy is maintained when the system is tested on data collected by using different methods. Spectrum accuracy is maintained in a patient sample that is, on average, more or less advanced in disease process or that has a somewhat different disease process or trajectory. Follow-up interval accuracy is maintained when the system is tested over a longer or shorter period.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Examples of receiver operating characteristic curves. Discrimination is the ability of the score to distinguish high-risk subjects from low-risk subjects. In the context of survival, it is usually assessed by Harrell c statistic, which is an extension of the area under the receiver operating characteristic curve. The c statistic is equal to the probability that a score will rank a high-risk subject higher than a low-risk subject. When the c statistic is >0.7, the score has acceptable discriminatory power. When the c statistic is equal to 0.5, the score has the same power as flipping a coin.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Examples of calibration curves. The calibration curve compares the predicted and observed probabilities of the event in different subgroups at risk (score strata). The global calibration performance can be tested with the Hosmer–Lemeshow goodness of fit test. A P value >0.05 suggests that the score is accurate on the whole.

Effect

After a tool has been developed with specific interventions and its accuracy and transportability have been assessed, one should evaluate how the utilization of this tool will affect clinical decision making and the care of patients (16). In other words, the tool has to be integrated into decision-making processes, and its benefit on health outcomes (such as HRQOL), patient satisfaction, and processes of care (such as improving clinical management pathways and tailoring individualized patient–centered strategies of care) must be shown.

Various epidemiologic intervention methodologies can be used, like cluster randomized, controlled trials (71–73) or controlled before-after studies. Advantages of cluster randomized, controlled trials over individually randomized, controlled trials include the ability to control for contamination across individuals, especially in studies that evaluate global strategies and not a specific intervention.

Dissemination

A dissemination strategy is a systematic plan for ensuring the dissemination of the prognostic tool to key internal and external stakeholders through diverse, effective, creative, and barrier-free methods after the tool has been finalized. The aim of the strategy is to ensure effective communication and dissemination of the prognostic tool to stakeholders to maximize use. Effect studies and tailored interventions can help this dissemination (74).

Figure 4 illustrates a framework for development of a new tool.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Framework for development of a new tool.