Quality-adjusted life-years as decisionmaker tools to address limitations inherent in number-needed-to-treat/harm values

One of the concerns that is often expressed by patient-outcomes researchers is how to get the results of cost–effectiveness analyses (CEA) implemented [1]. The problem stems partially from a disconnect between CEA results and relevance to actual clinical practice, as well as variations in CEA research quality [1–3]. Although researchers believe that the CEA results should change decision-makers’ viewpoints, what are the factors that interfere with full implementation? How can a practitioner be expected to make decisions using pharmacoeconomic data when relevance at the bedside is incomplete or even conflicting with clinical data? One of the problems may be the ambiguity associated with the use of quality-adjusted life-years (QALYs) as the outcome measure in CEA [4].

A commonly accepted measure of the benefit of a health treatment is the number needed to treat (NNT) to avoid one adverse health event (or to provide one cure). An analogous risk value is the number needed to harm (NNH); specifically, the number of patients treated that would result in one harmful side effect. These values are often reported in results of clinical trials and can also be calculated easily using data reported in clinical trial manuscripts. By comparing these values, an approximate benefit:risk assessment is constructed and then applied to clinical decision-making at the bedside or clinic.

The purpose of this article is to describe the potential relationships between NNT/NNH and QALYs. We apply an example of a clinical condition in which NNT and/or NNH values are similar. We hope that this discussion will help span the chasm between published CEA results and their clinical application.

Using NNT & NNH in clinical decisions

Most practicing clinicians lack a working understanding of pharmacoeconomic measures as they relate to treatment outcomes. Conversely, NNT is easily understood by the majority of clinicians and is widely used to assess the merits of a particular treatment and guide clinical decision-making [5]. Furthermore, treatments administered to a given patient need to be reviewed in terms of safety as well as efficacy. Therefore, the NNH is required to balance the relative risks with the benefits of a given treatment. The NNT and NNH are attractive measures to practicing clinicians owing to their simplicity; however, difficulties may arise when they are used to compare different drugs within or between therapeutic classes. This is largely due to different definitions of end points that are used in individual studies to define efficacy and safety.

Even in this sense, however, clinical decision-making is rather subjective in that clinicians may disagree regarding what constitutes an acceptable NNT:NNH ratio. In many instances, such as the administration of a HMG-CoA reductase inhibitor (statin) to a patient with hyperlipidemia and coronary artery disease, the benefit:risk ratio is widely regarded as favorable. The NNT is 12 (according to the Scandinavian Simvastatin Survival Study [4S]) to prevent a major cardiovascular event, whereas the NNH is comparable to placebo [6]. In other instances, however, the NNT and NNH have limited application in directing clinical decision-making, particularly in controversial situations in which one drug is not dominantly superior to the other.

A case in point is the use of glycoprotein IIb/IIIa receptor antagonists (GPRAs) for the treatment of non-ST segment elevation myocardial infarction (non-STEMI) in patients who are not scheduled to undergo percutaneous intervention (i.e., conservative strategy). The Platelet Glycoprotein IIb–IIIa in Unstable Angina: Receptor Suppression Using Integrilin Therapy (PURSUIT) and Platelet Receptor Inhibition in Ischaemic Syndrome Management in Patients Limited by Unstable Signs and Symptoms (PRISM-PLUS) studies investigated the safety and efficacy of eptifibatide and tirofiban, respectively, in patients presenting with non-STEMI [7,8]. However, the primary end points were different in the two studies: death or myocardial infarction (MI) at 30 days (PURSUIT), and death, MI or refractory ischemia at 7 days (PRISM-PLUS). The NNT for the primary end point in the PURSUIT study was 67 versus 19 in the PRISM-PLUS trial. While one certainly cannot determine the relative benefits of these two drugs based on these studies alone, many clinicians would consider refractory ischemia to be a ‘softer’ end point than death or MI, although NNT analyses weigh these individual components of the combined end point equally. Ultimately, despite these study and NNT differences, many clinicians consider these two drugs to be very similar in terms of efficacy for conservative management of non-STEMI. The PURSUIT study demonstrated a statistically significant benefit in terms of major adverse cardiovascular events in patients receiving eptifibatide compared with placebo but with an increased risk of major bleeds [7]. The NNT:NNH ratio was almost 1:1 with regards to these end points. A meta-analysis of the randomized trials summarized the data as NNT being 65 (95% CI: 40–203) for death or nonfatal infarct at 30 days [9]. The NNH of 150 for major bleeding fell within the confidence interval. Another meta-analysis summary of the research indicated the NNT to be 90 for MI or death and NNH to be 105 for major bleeding, with varying values by age group [10]. In the scenario where the values are so similar, do these values provide a clear decision-making tool or is the answer ambiguous? The clinician is in a conundrum whether to accept the cardiovascular benefits at the expense of an increased risk of bleeding. We also note that clinical practice guidelines, which are designed to assist in clinical decision-making, are not especially helpful in this particular instance, in which the recommendation is ‘it is reasonable to administer’ (Recommendation IIa) and the evidence is ‘limited’ (B) [11]. Thus, it is at the clinician’s discretion whether to prescribe this treatment.

There are other issues that complicate NNT and NNH analyses that should be, but sometimes are not, considered when making clinical decisions. For example, the costs of drugs are inherently missing from NNT and NNH analyses. This is especially important with high-cost drugs, such as GPRAs. This does not often weigh heavily to a practicing clinician but is watched closely by administrative personnel owing to its potential impact upon cost-containment strategies. We identified two research articles from clinical trial results describing the cost–effectiveness of GPRAs for the treatment of non-STEMI [12,13]. The first study was based upon results from PURSUIT and showed a small gain in QALY of 0.01 (0.84 vs 0.83; p = 0.45) in the aggressive medical management alone versus aggressive management plus eptifibitide. A summary of CEA research concluded that cost–effectiveness results were dependent upon risk stratification, that is, implementing treatment among patients who are most likely to benefit over time [14]. A health technology assessment review concluded that the CEA study results of GPRAs were suspect owing to the use of disease-specific end points rather than QALYs, differences in US versus UK health services, and an inability or failure to incorporate benefits beyond the study duration [12]. We also searched clinical guidelines for examples in which GPRAs were presented in terms of cost–effectiveness. We found no guidelines that provided clear guidance in this regard. Thus, CEA has yet to make a major impact in promoting the use of GPRAs for this indication.

Exploring the relevance of QALYs to clinical decision-making

Quality-adjusted life-years may encompass a more elaborate, yet practical, means of describing the benefit:risk ratio. We describe how QALYs weigh the variety of end points differently and may thus be a valuable tool, beyond NNT/NNH, to the clinician in therapeutic decision-making.

Quality-adjusted life-years were partly developed to avoid the problem of putting an economic value on life or an improvement in health. By applying a value-rating system, also known as a patient-preference measure (PPM), for a health condition, on a zero (i.e., death or worst possible health) to one (i.e., full or optimal health) scale, the value of health states across a variety of diseases can be compared [15]. Changes in PPMs captured over time are used to calculate QALYs by multiplying the amount of change by the time interval in years between measures.

How might QALYs be used to augment the clinical relevance of NNT and/or NNH and, thus, the applicability of pharmacoeconomic results? First, consider how NNT and NNH are captured within QALYs in the use of GPRAs for treatment of non-STEMI, with the prevention of death, MI or refractory ischemia as the benefit (NNT). Regarding harm (NNH), bleeding is the primary adverse outcome of interest. QALYs capture both the benefits and harm of each outcome. Assume that patients in a clinical trial of GPRA complete a PPM prior to and following the treatment to measure changes in PPMs that could be compared between the treatment and control groups. On the scale ranging from zero (i.e., death) to one (i.e., optimal health), the average PPM values may hypothetically look something like those displayed in Table 1. The NNT value primarily encompasses those patients in category one: ‘No MI, death, refractory ischemia or side effects’, compared with a combination of all other categories, and the NNH value primarily encompasses categories two, five and eight, compared with all other categories. PPMs and QALYs provide information regarding the impact of the different outcomes that are combined within NNT and/or NNH values. The total QALYs gained by treatment (active or placebo) is the sum (across all outcomes) of the proportion of patients who experience each outcome in Table 1 multiplied by the mean QALY gain (or loss) from that outcome. Therefore, the calculation of QALYs requires data regarding the rates of each outcome in addition to the impact of each outcome, whereas NNT and/or NNH summarizes those rates across several types of outcomes.

How well the QALY captures NNT/NNH, in the clinician’s view, may relate to the validity of the QALY measured in the condition under treatment, the method used to in collecting the values (within a clinical trial or through the literature), and the breadth of the population in which the data are captured (relationship to the clinician’s own patients). However, if one considers how NNT/NNH data are enfolded within QALYs, it may be more realistic to apply CEA in clinical decision-making.

We note that current controversies regarding QALYs impede their application in clinical decision-making. Two of those controversies are the facts that: the values used to develop PPM estimates are based on populations and QALYs are intended to be applicable across diseases [16,17]. Regarding the first, PPM instruments were developed based upon population perspectives because, in general, persons with a disease are likely to rate the value of that health state differently to those without the disease. If patients with the disease rate their condition higher than the population, there would be less potential gain by improving the condition, thereby biasing health decisions against treating the condition [17]. Therefore, application of PPMs to individual patients in a clinical situation requires consideration of this limitation. Concerning the second controversy, PPMs and QALYs were developed to compare outcomes across diseases so that CEA could be used to compare many treatments within health systems and direct policy decisions. PPMs are, by nature, general measures of health-related quality of life (HRQoL) [16]. Thus, instruments used to measure PPMs may not be sensitive to HRQoL changes that are more specific to a particular disease. Therefore, it is important to recognize that QALYs are one facet of the clinical decision-making process. These controversies should be considered but do not preclude the use of QALYs in conjunction with NNT/NNH. What is required, however, is a theoretically sound and practically acceptable measure of PPMs captured within efficacy or effectiveness studies.

Consider one of the most basic measures of PPM, the feeling thermometer (FT) [18]. The FT is administered by showing patients a visual analog scale ranging from zero to one and asking patients to rate their current HRQoL by marking on that scale. This tool can be used at clinical trial time points or during patient–provider encounters to gauge how much each individual’s health has changed, either positively or negatively, since the last visit. This is similar to the pain scales that are commonly administered to patients in preparation for a provider encounter. Based upon the outcome being experienced by the patient, the FT could capture its impact. We note that a weakness of the FT as a sole measure of PPM is that it is not based on choice, and patients tend to weight an acute illness over a chronic condition disproportionately [18]. However, it may be helpful for use in conjunction with other PPM or HRQoL instruments to provide an immediate visual measure of the impact of an outcome on the patient. We note that the EuroQoL Five Domains survey instrument includes the FT in addition to the HRQoL questions that are used to determine PPMs [19].

Conclusion

Reliance upon NNT/NNH as the primary clinical decision tool is flawed owing to its lack of consideration of the impact of the outcomes on the patients’ HRQoL measures and use of composite end points to calculate these values. Although not ideal, QALYs provide more information regarding the relative impact of different outcomes that may result from therapeutic alternatives. Furthermore, NNT/NNH data are incorporated into QALYs because the proportion of patients experiencing each outcome is used to calculate the total QALYs gained (or lost) from a treatment. In this era of limited healthcare resources, clinicians, as well as policymakers, require tools that go beyond NNT and/or NNH values. Comparative efficacy and effectiveness trials that capture PPMs and QALYs are needed.

Table 1. Hypothetical patient-preference measures and quality-adjusted life-years associated with outcomes from conservative treatment of non-ST-elevation acute coronary syndromes using glycoprotein IIb/IIIa receptor antagonists.

Category of events experienced by patients	Mean PPM value	Mean PPM gain/loss per person at 1 month^*	QALY (1-month PPM change/12)
Baseline	0.50
No MI, death, refractory ischemia or side effects (i.e., the greatest potential QALY gain)	0.84	0.34	0.028
No MI, death or refractory ischemia but major bleed	0.70	0.20	0.017
No MI, death, refractory ischemia or major bleed but other side effects	0.75	0.25	0.021
MI but no side effects	0.40	-0.10	-0.008
MI and major bleed	0.30	-0.20	-0.017
MI and no major bleed but other side effects	0.35	-0.15	-0.013
Refractory ischemia but no side effects	0.50	0.00	0.000
Refractory ischemia and major bleed	0.40	-0.10	-0.008
Refractory ischemia and no major bleed but other side effects	0.45	-0.05	-0.004
Death (i.e., the greatest QALY loss)	0.00	-0.50	-0.042

^*Change from baseline.

MI: Myocardial infarction; PPM: Patient-preference measure; QALY: Quality-adjusted life-year.

Acknowledgements

The authors acknowledge the assistance of Prasad Bhave, MD in reviewing the literature cited in this manuscript.

Financial & competing interests disclosure

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.