SEARCH

SEARCH BY CITATION

Keywords:

  • decision models;
  • health economics methods;
  • health surveys;
  • health-state utility;
  • methodology

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

Background:  The methods used to estimate health-state utility values (HSUV) for multiple health conditions can produce very different values. Economic results generated using baselines of perfect health are not comparable with those generated using baselines adjusted to reflect the HSUVs associated with the health condition. Despite this, there is no guidance on the preferred techniques and little research describing the effect on cost per quality adjusted life-year (QALY) results when using the different methods.

Methods:  Using a cardiovascular disease (CVD) model and cost per QALY thresholds, we assess the consequence of using different baseline health-state utility profiles (perfect health, no history of CVD, general population) in conjunction with models (minimum, additive, multiplicative) frequently used to approximate scores for health states with multiple health conditions. HSUVs are calculated using the EQ-5D UK preference-based algorithm.

Results:  Assuming a baseline of perfect health ignores the natural decline in quality of life associated with age, overestimating the benefits of treatment. The results generated using baselines from the general population are comparable to those obtained using baselines from individuals with no history of CVD. The minimum model biases results in favor of younger-aged cohorts. The additive and multiplicative models give similar results.

Conclusion:  Although further research in additional health conditions is required to support our findings, our results highlight the need for analysts to conform to an agreed reference case. We demonstrate that in CVD, if data are not available from individuals without the health condition, HSUVs from the general population provide a reasonable approximation.


Introduction

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

A number of agencies, including the National Institute for Health and Clinical Excellence (NICE), require economic evidence to be presented in the form of cost-effectiveness analyses whereby health benefits are quantified by quality adjusted life-years (QALYs) [1]. QALYs are calculated by summing the time spent in a health state weighted by the health-state utility value (HSUV) associated with the health state, thus incorporating both length of survival and HSUVs into a single metric. Classification systems can produce a wide range of values for the same health state and the economic results generated using different systems are not always comparable [1]. Consequently, for submissions in the UK, the Institute advocate a preference for EQ-5D data with HSUVs obtained using UK population weights when available [1].

However, this is not sufficient to ensure consistency across appraisals because there is no guidance on appropriate baseline HSUVs that should be used to quantify the underlying health condition for patients entering the model [1]. If a baseline utility of perfect health (i.e., EQ-5D equals 1) is used to represent the absence of a health condition, the incremental QALYs gained by an intervention are inflated [2] and the results obtained using a baseline of perfect health are not comparable with those obtained when the baseline is adjusted for not having a particular health condition [3]. There is currently no consensus on baseline HSUVs used in economic evaluations.

In addition, there is currently no directive on the method that should be used to combine HSUVs for multiple health conditions. Analysts are increasingly exploring the benefits of interventions in individuals with several comorbid conditions. For example, HMG-CoA reductase inhibitors (statins) reduce both cardiovascular (CV) risk and rheumatoid arthritis (RA) disease activity; and an economic model exploring the benefits of statins in this population would include health states for patients with a history of both RA and cardiovascular disease (CVD) [4]. Because of strict exclusion criteria preventing patients with comorbidities entering clinical trials, it is unlikely that HSUVs will be available from patients with both health conditions.

When HSUVs for the multiple health states are not available, approximate scores are estimated by combining data collected from patients with the individual health conditions. Three methods are frequently used: 1) additive; 2) multiplicative; and 3) minimum models. The additive and multiplicative models assume a constant absolute or proportional effect, respectively, while the minimum model applies a disutility that can vary depending on the baseline utility modeled. Research exploring the appropriateness of the techniques used to combine utility values is inconclusive. The additive and multiplicative models have been shown to produce similar results for individuals with both diabetes and thyroiditis [5]; the multiplicative model produced accurate utilities for several comorbid conditions [6]; and the minimum model was advocated as the preferred methodology in two other studies [7,8].

Although literature describing minimum requirements for probabilistic analyses is growing [9], research exploring the basic principles involved in using HSUVs in economic models, and the implications for results generated from the models when using different techniques is scarce. The limited research undertaken in this area has explored the appropriateness of different baseline utilities and approximate HSUVs for multiple health conditions in isolation; and there is currently no consensus on the preferred methodologies when the two adjustments are undertaken together.

We describe the results of a pilot study in which we explore the effect of using different baseline utility values and different techniques to estimate approximate HSUVs for multiple health conditions in combination. We use an existing economic model and data from the Health Survey for England to investigate the potential effect on policy decision-making using cost per QALY thresholds. The primary objective of the study is to instigate additional research in this area to provide a foundation for better practice in economic evaluations used to inform health care decision-makers in the UK and elsewhere.

Methods

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

The following section provides a brief description of the economic model and a synopsis of the data used.

Cardiovascular Model

An existing peer-reviewed Markov model [10] was modified slightly so that the health states (Fig. 1) matched the definitions of three CV conditions available from the Health Survey for England which are angina (A), heart attack (HA) and stroke (Str) [10,11]. The model compares two alternative treatments and an annual cycle is used for transitions between health states. Individuals enter the model in the event-free (EF) health state and can move to a primary health state: angina (A), nonfatal heart attack (HA), or nonfatal stroke (Str), or remain in the EF health state. Individuals in the primary and post-event health states can move to a subsequent health state: subsequent angina (SA), subsequent nonfatal HA (SHA), subsequent nonfatal stroke (SStr); or remain in the primary or post-event health state. In each cycle all individuals are at risk of death through other causes (DoC), or fatal CVD (fCVD). Health-state costs are taken from a recent HTA evaluation of lipid treatments in the UK [10].

image

Figure 1. Health states in cardiovascular model.

Download figure to PowerPoint

Health Survey for England

The Health Survey for England (HSE) is conducted annually using random samples of the population living in private households in England. The 2003 and 2006 surveys included questions about history of CVD and a random sample of participants (aged 16 to 98 years) were asked to complete the EQ-5D questionnaire (N = 26,679) [11,12]. Preference-based HSUVs were estimated using the weights obtained using time trade off valuations from the UK general public [13].

We assumed that the data from individuals who reported a history of just one CV condition are representative of the HSUVs of individuals who have a first ever primary CV event; and that data from individuals who reported a history of more than one CV condition are representative of the HSUVs of individuals who have a subsequent event (Table 1). For example, the mean HSUV during the first 12 months after experiencing a primary (secondary) heart attack is 0.721 (0.431) and the corresponding mean HSUV for time periods after this is 0.742 (0.685).

Table 1.  EQ-5D scores sub\grouped by health condition and time since event
Health conditionHealth stateNAgeEQ-5DSE
MeanMean
Utility values used to populate health states in the economic model
Event-freeEF25,08047.00.8720.001
Angina <12 months, history of just anginaA27168.80.6150.019
No event <12 months, history of just anginapA24668.00.7750.015
Angina <12 months, history of angina + other CV conditionSA24567.90.5410.022
No event <12 months, history of angina + other CV conditionpSA18469.40.7150.022
Heart attack <12 months, history of just heart attackHA3165.40.7210.045
No event <12 months, history of just heart attackpHA20665.10.7420.020
Heart attack <12 months, history of heart attack + other CV conditionSHA3666.70.4310.066
No event <12 months, history of heart attack + other CV conditionpSHA18469.20.6850.024
Stroke <12 months, history of just strokeStr7667.90.6260.038
No event <12 months, history of just strokepStr29166.80.6680.018
Stroke <12 months, history of stroke + other CV conditionSStr1873.50.4790.087
No event <12 months, history of stroke + other CV conditionpSStr7770.40.6410.037
Data used to compare methods for estimating proxy scores for multiple health conditions
Angina (t = ever), history of just angina 51768.40.6910.013
Heart attack (t = ever), history of just heart attack 23766.60.7390.018
Stroke (t = ever), history of just stroke 36767.00.6600.016
Angina and heart attack (t = ever) 32368.20.6240.019
Angina and stroke (t = ever) 6370.30.5960.043
Heart attack and stroke (t = ever) 3269.70.5380.065
Angina <12 months and heart attack <12 months 2363.10.4000.073
Angina <12 months and heart attack >12 months 15468.40.5850.030

The relationship between HSUVs, age, sex, and history of CVD was explored using ordinary least square regressions. Model 1 (EQ-5D = 0.9508566 + 0.0212126*male − 0.0002587*age − 0.0000332*age2, Fig. 2) was obtained using the full dataset (n = 26,679) and can be used to estimate the mean HSUVs for individuals in the general population. Model 2 (EQ-5D = 0.9454933 + 0.0256466*male − 0.0002213*age − 0.0000294*age2, Fig. 2) was obtained from individuals who reported no history of angina, heart attack or stroke (n = 25,080) and can be used to estimate the HSUVs for individuals with no history of CVD [14].

image

Figure 2. Baseline utility for the event-free health state: Relationship between HSUVs, age, sex, and history of CVD.

Download figure to PowerPoint

Analyses

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

The following section describes a worked example demonstrating the difference in incremental QALYs gained from avoiding a single event when using different baseline HSUV profiles, followed by results generated from the economic model demonstrating the potential effect on a policy decision using a cost per QALY threshold when using the different baseline HSUV profiles. We then provide a worked example using the three alternative models to estimate approximate scores for multiple health conditions, looking at the difference in incremental QALYs associated with avoiding a single event, followed by results generated from the economic model when combining the different baseline profiles and the techniques used to combine the utility data.

Baseline HSUV Profiles

In a CV model, individuals who are at high risk of a CV event and have no prior history of CVD typically enter the model in an “event-free” health state. The HSUV profile associated with this health state is then used as the baseline to estimate the health benefits accrued through avoiding CV events. Ideally, the health profile for the EF health state would be derived from long term registry data and would represent the HSUVs for individuals who are at high risk of a primary CV event but who have no existing history of CVD. In the absence of these data, analysts assume the baseline HSUV profile is either 1) equal to perfect health (i.e., EQ-5D = 1 irrespective of age or sex), 2) equal to the profile of HSUVs from the general population adjusted for age and sex (i.e., all individuals irrespective of history of CVD), or 3) equal to the profile of HSUVs from individuals with no history of CVD. To simplify the methodology, we present examples for males only in the following section.

In the following example (Box 1), we illustrate the difference in QALYs accrued from avoiding a single event using the three alternative baseline HSUV profiles for the EF health state. The HSUV profile when assuming a baseline of perfect health (UPHEF) is constant at EQ-5D = 1. The HSUV profile when assuming a baseline from the individuals with no history of CVD (UNCVEF) is calculated using Model 2 and the HSUV profile when assuming a baseline from the general population (UGPEF) is calculated using Model 1 (Fig. 2). The mean EQ-5D score for individuals who reported experiencing angina within the previous 12 months (UA) is 0.6148 and the mean age for this subgroup is 68.8 years (Table 1). We assume the event occurs at the age of 50 years and examine the cumulative and incremental QALYs accrued over a 50-year time horizon. For the examples using the age-adjusted baseline profiles, the data for the individual health conditions are combined multiplicatively (see Box 2 for more details on this technique).

Box 1

Comparing the incremental QALY gain from a single event when using different baseline HSUV profiles

Let UPHEF = 1

UNCVEF = 0.9454933 + 0.0256466 * male − 0.0002213 * age − 0.0000294 * age2

UGPEF = 0.9508566 + 0.0212126 * male − 0.0002587 * age − 0.0000332 * age2

UA = 0.6148 (mean age = 68.8 years)

Where Uij = HSUV, and i = baseline: PH = perfect health

     NCV = no history of CVD (regression Model 2)

     GP = general population (regression Model 1)

   j = health state: EF = event-free, A = angina

multiplier for angina for UNCV: male = 0.753 (= 0.6148/0.8167)

multiplier# for angina for UGP: male = 0.771 (= 0.6148/0.7973)

(# see example 2 for method used to obtain multipliers)

Results when assuming a baseline HSUV profile of full health:

 Cumulative QALYPHEF = 50

 Cumulative QALYPHA = 30.74

 Incremental QALYPH = QALYPHEF − QALYPHA = 19.26

Results when using a baseline HSUV profile from individuals with no history of CVD:

 Cumulative QALYNCVEF = 39.27

 Cumulative QALYNCVA = 29.56

 Incremental QALYNCV = QALYNCVEF − QALYNCVA = 9.71

Results when assuming a baseline HSUV profile from the general population:

 Cumulative QALYGPEF = 38.08

 Cumulative QALYGPA = 29.37

 Incremental QALYGP = QALYGPEF − QALYGPA = 8.71

Box 2

Estimating an approximate HSUV for the multiple health state both angina and heart attack

Let j = health state and: AHA = both angina and heart attack, A = angina, HA = heart attack, A,HA = approximate angina plus heart attack

 δij = disutility; ϕij = multiplier; min = minimum

 UAHA @ mean age 68.2 = 0.6243, UGP @ age 68.2 = 0.8000, UNCV @ age 68.2 = 0.8193

 UA @ mean age 68.4 = 0.6910, UGP @ age 68.4 = 0.7990, UNCV @ age 68.4 = 0.8185

 UHA @ mean age 66.6 = 0.7391, UGP @ age 66.6 = 0.8076, UNCV @ age 66.6 = 0.8260

Using a baseline HSUV profile from individuals with no history of CVD,

Additive: δNCVAHA = UNCV − UAHA = 0.8193 − 0.6243 = 0.1950

  δNCVA,HA = δNCVA + δNCVHA = (UNCV − UA) + (UNCV − UHA)

   = (0.8185 − 0.6910) + (0.8260 − 0.7391) = 0.2143

Multiplicative: ϕNCVAHA = UAHA/UNCV = 0.6243/0.8193 = 0.7622

  ϕNCVA, HA = ϕNCVANCVHA = (UA/UNCV)*(UHA/UNCV)

   = (0.6910/0.8185)*(0.7391/0.8260) = 0.7555

Minimum: UNCVAHA = min(UNCV,UAHA) = min(UNCV,0.6243)

  UNCVA,HA = min(UNCV,UA,UHA) = min(UNCV,0.6910,0.7391)

Assuming the event occurs at the age of 50 years,

Using the data from individuals with a history of both angina and heart attack:

 Additive, incremental QALYsNCV = QALYNCVEF − QALYNCVAHA = 39.27 − 29.52 = 9.75

 Multiplicative, incremental QALYsNCV = QALYNCVEF − QALYNCVAHA = 39.27 − 29.92 = 9.35

 Minimum, incremental QALYsNCV = QALYNCVEF − QALYNCVAHA = 39.27 − 31.22 = 8.05

Using the approximate scores from individuals with a history of either angina or heart attack:

 Additive, incremental QALYsNCV = QALYNCVEF − QALYNCVA,HA = 39.27 − 28.55 = 10.72

 Multiplicative, incremental QALYsNCV = QALYNCVEF − QALYNCVA,HA = 39.27 − 29.67 = 9.60

 Minimum, incremental QALYsNCV = QALYNCVEF − QALYNCVA,HA = 39.27 − 34.46 = 4.81

When using a baseline HSUV profile from the general population, the approximate HSUVs are calculated using the same method as above replacing the values from individuals with no history of CVD with the corresponding values from the general population. The results are provided in Table 4. N.B. any anomalies in the results are caused by rounding in the decimal places in the calculations shown above.

Table 4.  Cumulative and incremental QALY gains from a single event using different techniques to estimate proxy scores for multiple health states
 Cumulative QALYIncremental QALYError in Incremental QALY
Observed*ProxyBaseline – observedBaseline – proxy
GPNCVDGPNCVDGPNCVDGPNCVDGPNCVD
  • *

    Using utility data from individuals with a history of both conditions;

  • Using data from individuals with a history of a single condition to estimate the HSUV for the multiple health condition.

  • GP, general population; NCVD, No history of CVD.

Baseline
Event-free38.139.3        
Angina plus Heart Attack: EQ-5D just angina = 0.691; EQ-5D just heart attack = 0.739; EQ-5D angina plus heart attack = 0.624
Additive29.329.529.228.58.89.88.810.70.01.0
Multiplicative29.729.930.129.78.49.47.99.6−0.40.2
Minimum31.231.234.134.56.98.13.94.8−2.9−3.3
Angina plus Stroke: EQ-5D just angina = 0.691; EQ-5D just stroke = 0.660, EQ-5D angina plus stroke = 0.596
Additive28.428.525.424.79.710.712.714.63.03.9
Multiplicative28.728.927.026.59.410.411.112.71.82.3
Minimum29.829.832.933.08.39.55.26.3−3.1−3.2
Heart Attack plus Stroke: EQ-5D just heart attack = 0.739, EQ-5D just stroke = 0.660; EQ-5D heart attack plus stroke = 0.538
Additive25.325.527.426.712.813.810.712.6−2.0−1.2
Multiplicative25.826.028.628.112.213.39.511.1−2.7−2.1
Minimum26.926.932.933.011.212.45.26.3−6.0−6.1
Angina <12 months, Heart Attack <12 months:
EQ-5D angina <12 months = 0.615; EQ-5D heart attack <12 months = 0.721; EQ-5D angina <12 months plus heart attack <12 months = 0.400
Additive16.917.324.423.721.222.013.715.6−7.5−6.4
Multiplicative18.518.726.125.719.620.612.013.6−7.6−7.0
Minimum20.020.030.830.818.119.37.38.5−10.8−10.8
Angina <12 months, Heart Attack >12 months:
EQ-5D angina <12 months = 0.615; EQ-5D heart attack >12 months = 0.742; EQ-5D angina <12 months plus heart attack >12 months = 0.585
Additive27.427.625.424.710.711.712.714.62.02.9
Multiplicative27.928.126.826.410.211.211.312.91.11.7
Minimum29.329.330.830.88.810.07.38.5−1.5−1.5

The cumulative QALYs for the EF health state are calculated by summing the life-years weighted by the HSUV profile across the 50 year period (Cumulative QALYPHEF = 50*1, Cumulative inline image, Cumulative inline image). The cumulative QALYs for angina are calculated by summing the life-years weighted by the baseline profile multiplied by the multiplier associated with angina (Cumulative QALYPHA = 50*1*0.6148 = 30.74, Cumulative inline image, Cumulative inline image. The incremental QALYs associated with avoiding angina is calculated as the difference between the total cumulative QALYs for the EF health state minus the total cumulative QALYs for angina (Cumulative QALYiEF − Cumulative QALYA). The technique used to obtain the multipliers is described in the next worked example.

Comparing results when using different baseline HSUV profiles for the EF health state.  For a male, the cumulative QALYs (Box 1) associated with remaining in the EF health state range from 38.1 when using a baseline HSUV profile from the general population to 50 when using a baseline HSUV profile of perfect health; and the cumulative QALYs associated with angina range from 29.4 when using a baseline HSUV profile from the general population to 30.7 when using a baseline HSUV profile of perfect health. The incremental QALY gain associated with avoiding angina range from 8.71 when using a baseline HSUV profile from the general population to 19.26 when using a baseline HSUV profile of perfect health. The incremental QALYs obtained using the baseline HSUV profile from the general population are comparable to those obtained when using the baseline HSUV profile from individuals with no history of CVD (8.71 vs. 9.71).

Looking at the QALY gain associated with avoiding a single heart attack or a stroke (Table 2), the values obtained when assuming a baseline HSUV profile of perfect health are substantially higher than those obtained using the age adjusted data. Again, the QALY gain obtained using the baseline HSUV profile from the general population are comparable to those obtained using the baseline HSUV profile from individuals with no history of CVD (heart attack: 4.30 vs. 5.18; stroke: 8.33 vs. 9.30).

Table 2.  Cumulative and incremental QALYs associated with a single event using different baseline utility data
 Multiplicative Model
Cumulative QALYIncremental QALY
Baseline: perfect health
Event-free50.00 
Angina30.7419.26
Heart Attack36.0713.94
Stroke31.3118.69
Baseline: from general population  
Event-free38.08 
Angina29.378.71
Heart attack33.784.30
Stroke29.758.33
Baseline: from individuals with no history of CVD
Event-free39.27 
Angina29.569.71
Heart attack34.095.18
Stroke29.979.30

Cost per QALY Results Using Different Baseline HSUV Profiles for the EF Health State

The three alternative baseline profiles were applied in the CVD model and used to assess the lifetime benefits associated with avoiding primary events for cohorts of differing ages (Table 3). The results from the worked example show the benefits associated with avoiding a single event are considerably larger when using a baseline of perfect health compared to adjusting the baseline. When examining the effect on the results generated from the model, the cost per QALY obtained using a baseline of perfect health (Fig. 3) is substantially lower than the corresponding results obtained using the age-adjusted profiles, particularly for the older aged cohorts. If a threshold of £20,000 per QALY is applied (Fig. 3), using a baseline of perfect health could potentially induce a different policy decision than the one based on results generated when using a baseline HSUV profile that is adjusted for not having the health condition.

Table 3.  Results generated from CVD model using the three alternative baseline profiles (combining utility scores multiplicatively)
 Baseline utilityTreatment A QALYsTreatment B QALYsIncremental QALYsCost per QALY
Age 50 years
Costs £(,000)£4,216£5,610£1,394 
 Perfect health16,79516,895100£13,887
General population14,12914,17849£28,324
No history of CVD14,36314,41754£25,914
Age 60 years
Costs £(,000)£3,660£4,773£1,113 
QALYsPerfect health13,58213,64867£16,711
General population10,91910,95233£33,957
No history of CVD11,19711,22932£34,777
Age 70 years
Costs £(,000)£2,609£3,424£815 
QALYsPerfect health9,96610,00236£22,849
General population7,6437,65613£62,195
No history of CVD7,8667,88014£56,487
image

Figure 3. Comparing the results generated from the CVD model using the three alternative baseline profiles.

Download figure to PowerPoint

Estimating Approximate HSUV for Multiple Health Conditions

In the following example (Box 2), we use data from individuals who have a history of angina and no other CV condition (UA) and data from individuals who have a history of a heart attack and no other CV condition (UHA) to estimate an HSUV for the multiple health state “angina and heart attack” (UA,HA). The additive, multiplicative and minimum models are used to estimate the HSUV profiles for the multiple health condition using the disutility (δij), multiplier (ϕij) or minimum value (min), respectively. These are then used in conjunction with the two age-adjusted baseline HSUV profiles (no history of CVD and general population) but not the baseline of perfect health. We compare the QALYs obtained from avoiding a single event when using the HSUV (UAHA) from individuals who have a history of both angina and a heart attack with those obtained when using the estimated HSUV (UA,HA).

A. Using the HSUV obtained from individuals with a history of both angina and heart attack.  The mean HSUV for individuals with a history of both angina and heart attack (UAHA) is 0.6243, and the mean age for this subgroup is 68.2 years. When using the baseline HSUV profile from the general population, the HSUV for a male at the age of 68.2 years (UGP) is 0.8000 (from Model 1). For the additive model, the disutility (δGPAHA) is the absolute difference between the baseline utility at the age of 68.2 years and the HSUV associated with the health condition angina and heart attack (i.e., δGPAHA = UGP-UAHA = 0.8000 − 0.6243 = 0.1757). When summing the QALYs accumulated for the health condition, as the additive model assigns a constant effect irrespective of age, a constant value of 0.1757 is deducted from the age-adjusted baseline HSUV each year and the resulting values are summed to give the total cumulative QALYs (inline image). The incremental QALYs are then calculated by deducting the total cumulative QALYs associated with the condition angina and heart attack (QALYGPAHA = 29.30) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08).

For the multiplicative model, the multiplier (ϕGPAHA) is the value that will give the HSUV associated with the health condition angina and heart attack (UAHA) when multiplying the baseline utility at the age of 68.2 years (i.e., ϕGPAHA = UAHA/UGP = 0.6243/0.8000 = 0.7804). When summing the QALYs accumulated for the health condition, the multiplicative model assigns a constant proportional effect which is dependent on the age-adjusted baseline HSUV. The total cumulative QALYs are calculated by summing the QALYs obtained when multiplying the age-adjusted baseline HSUV with the corresponding multiplier (inline image). The incremental QALYs are then calculated by deducting the total cumulative QALYs associated with the condition angina and heart attack (QALYGPAHA = 29.72) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08). For the minimum model, the minimum HSUV for the multiple condition angina and heart attack, and the age-adjusted baseline is used. Consequently, the detriment associated with the health condition angina plus heart attack is not constant. The total cumulative QALYs is simply the sum of the minimum value each year (inline image). The incremental QALYs are then calculated by deducting the total cumulative QALYs for the health-state angina plus heart attack (QALYGPAHA = 31.21) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08).

B. Using the HSUV obtained from individuals with a history of either angina (with no other CV condition) or heart attack (with no other CV condition).  The mean HSUV for individuals with a history of just angina (UA) is 0.6910 and the mean HSUV for individuals with a history of just heart attack (UHA) is 0.7391. The mean ages for these subgroups are 68.4 and 66.6 years, respectively. When using the baseline HSUV profile from the general population, the corresponding HSUVs for a male at the age of 68.4 and 66.6 years are 0.7990 and 0.8076 (from Model 1). For the additive model, the total disutility (δGPA,HA) is estimated to be the sum of the absolute difference between the baseline utility at the age of 68.4 and the HSUV associated with the health condition angina (i.e., δGPA = UGP-UA = 0.7990 − 0.6910 = 0.1080) plus the absolute difference between the baseline utility at the age of 68.4 and the HSUV associated with the health condition heart attack (i.e., δGPHA = UGP-UHA = 0.8076 − 0.7391 = 0.0686), giving a total estimated detriment of 0.1766. When summing the QALYs accumulated for the health condition, a constant value of 0.1766 is deducted from the age-adjusted baseline HSUV each year and the resulting values are summed to give the total cumulative QALYs (inline image). The incremental QALYs are then calculated by deducting the total cumulative QALYs (QALYGPA,HA = 29.25) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08).

For the multiplicative model, the estimated multiplier for the health-state angina and heart attack (ϕGPA,HA) is calculated by multiplying the multiplier for angina (ϕGPA) with the multiplier for heart attack (ϕGPA). The single multipliers are calculated using the method described earlier, that is, the multiplier for angina is obtained using the HSUV for angina and the baseline HSUV for individuals at the age of 68.4 years (ϕGPA = 0.6910/0.7790) and the multiplier for heart attack is obtained using the HSUV for heart attack and the baseline HSUV for individuals at the age of 66.6 years (ϕGPA = 0.7391/0.8076). When multiplied together, the estimated multiplier for the combined conditions angina and heart attack (ϕGPA,HA) is 0.7913. The total cumulative QALYs are calculated by summing the QALYs obtained when multiplying the age-adjusted baseline HSUV with the corresponding multiplier (inline image). The incremental QALYs are then calculated by deducting the total cumulative QALYs associated with the condition angina and heart attack (QALYGPA,HA = 30.13) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08).

For the minimum model, the minimum HSUV for the individual conditions angina and heart attack, and the age-adjusted baseline is used. The total cumulative QALYs is simply the sum of the minimum value each year (inline image). The incremental QALYs are then calculated by deducting the estimated total cumulative QALYs for the health-state angina plus heart attack (QALYGPA,HA = 34.14) from the baseline total cumulative QALYs for the EF health state (QALYGPEF = 38.08).

Comparing Results When Estimating Approximate HSUVs for Multiple Health Conditions

When using age-adjusted baseline utilities from the general population to represent the HSUV for the EF health state, and the HSUV for individuals with a history of both angina and heart attack, the incremental QALYs obtained using the additive and the multiplicative models are 8.79 and 8.36, respectively, compared with 6.87 when using the minimum model. The corresponding incremental QALYs obtained when estimating HSUVs for the combined health state are 8.83, 7.95, and 3.94 for the additive, multiplicative, and minimum models, respectively. If it is assumed that the values obtained using the data from individuals with both health conditions are correct, then the additive and multiplicative models produce much smaller errors in the incremental values than the minimum model.

Using age-adjusted baseline utilities from individuals with no history of CVD to represent the HSUV profile for the EF health state (calculations provided in Box 2), the additive and the multiplicative models again produce similar results with incremental QALYs of 10.72 and 9.60, respectively, compared with 9.75 and 9.35 when using the data from individuals with a history of both conditions. The incremental QALY gain when using the minimum model is much smaller at 4.81 and 8.05 when using the HSUV from the individual health conditions and the HSUV from individuals with both health conditions, respectively. Results for additional examples (n ≥ 20) are provided in Table 4.

Cost per QALY Results Generated When Combining the Alternative Baseline HSUV Profiles with the Three Different Models Available to Combine HSUVs

The three alternative techniques used to combine utility scores are applied in the CVD model and used to assess the lifetime benefits associated with avoiding primary events for cohorts of differing ages using a baseline from individuals with no history of CVD and a baseline from individuals from the general population (Table 5). The results from the second worked example showed the benefits associated with avoiding a single event are considerably smaller when using the minimum model to combine the utility values. This has a larger effect on the results for older aged cohorts (Table 5) where the ratio of costs and QALYs are more sensitive to small differences in the number of incremental QALYs gained. Figure 4 shows the cost per QALY results generated from the model using the different techniques to combine the utility data. There is very little difference in the results for the additive and multiplicative models, with the baseline HSUVs having a larger effect than the technique used to combine the utility data.

Table 5.  Results generated from the CVD model when combining different baseline utility scores and different methods to combine utility data
 General populationNo history of CVD
AdditiveMultiplicativeMinimumAdditiveMultiplicativeMinimum
Age 55 years
Treatment A, total QALY12,53012,53512,56512,79012,79412,827
Treatment B, total QALY12,57312,57712,60512,83712,84112,870
Incremental QALY434340474743
Cost per QALY£29,109£29,394£31,742£26,664£26,927£29,088
Age 65 years
Treatment A, total QALY9,2579,2629,2989,5109,5159,553
Treatment B, total QALY9,2829,2869,3189,5379,5429,576
Incremental QALY252420272723
Cost per QALY£38,680£39,553£47,253£35,235£36,021£42,767
Age 75 years
Treatment A, total QALY6,0386,0426,0676,2516,2566,284
Treatment B, total QALY6,0496,0536,0756,2646,2686,293
Incremental QALY1111813129
Cost per QALY£58,521£61,078£82,287£52,676£54,892£74,144
image

Figure 4. Comparing results generated from the CVD model when combining different baseline utility scores and different methods to combine utility data.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

We have demonstrated that the difference in QALY benefits accrued from avoiding a single CV event when using a baseline of perfect health are not comparable with those accrued when using a baseline that is adjusted for not having CVD. We have also demonstrated that in CVD, results generated using age-adjusted data from the general population are comparable to those obtained using a baseline from individuals with no history of CVD. Applying the different approaches in an economic model, we also show that assuming an HSUV profile of perfect health as the baseline could potentially influence a policy decision based on a cost per QALY threshold.

The HSE data show that both age and sex are independent predictors of HSUVs and these findings are observed in numerous other datasets [13,16]. Given that the mean EQ-5D score is never equal to full health irrespective of age or sex, using a baseline of perfect health overestimates the benefits associated with avoiding an event and biases the results in favor of the older age cohorts because it ignores the natural decline in mean HSUVs because of age and comorbidities. Data obtained from individuals without the health condition under consideration is the ideal baseline profile and should be used where possible. However, if these data are not available, we show that in CVD, the results generated using age-adjusted baseline data from the general population are comparable with the results generated using age-adjusted baseline data from individuals with no history of CVD.

It should be noted that for the first example where we explore the effect on the ICER of using different baseline profiles we combine the data for the age-adjusted analyses using the multiplicative model. We could have combined the data additively or used the minimum model but felt that presenting all three sets of results added an unnecessary complexity to the methods and detracted from the purpose of the exercise which was twofold. First, to reiterate findings previously described by Flanagan, i.e., that using a baseline of perfect health over estimates the benefits of treatment, and second, to take the research one step further by exploring the potential effect on policy decisions using results generated from an economic model.

We demonstrated that when combined with the age-adjusted utilities, the method used to estimate approximate scores for multiple health conditions can produce a large variation in the incremental QALY gain from avoiding a single event. When applying the techniques in the economic model we demonstrate that the method used to estimate the approximate scores could affect a policy decision based on a cost per QALY threshold. In particular, using the minimum model in combination with an age-adjusted baseline produces results that are not comparable to those generated using the additive or multiplicative models.

The existing literature describing the effect on results when combining HSUVs using different methods is sparse and inconclusive. Both Dale et al. and Fu et al. suggest the minimum value should be used to approximate the HSUV for a multiple health condition [7,8]. By taking the minimum mean utility score of the individual health conditions that contribute to a multiple health condition, the minimum model assumes that comorbidity has no additional detrimental effect on the HSUV of individuals with an existing health condition. This is counterintuitive and data from the HSE show that, in CVD, there is a statistically significant difference in the mean EQ-5D score for individuals with one condition compared with those with more than one CV condition (mean EQ-5D for individuals with a history of just angina = 0.691, mean EQ-5D for individuals with a history of angina and stroke = 0.596, P < 0.01). In addition, when applying the minimum model in an economic model in conjunction with an age-adjusted baseline, the method fails. The HSUVs for individuals who experience a primary heart attack is 0.7213. In the primary prevention analyses where all individuals commence in the EF health state the age-adjusted EQ-5D score for males with no history of CVD at the age of 89 years is 0.718. Consequently, when using the minimum model there is no benefit in avoiding a non fatal heart attack in males over the age of 89 years. Similarly, the post primary angina health state has a mean EQ-5D score of 0.775; thus, there are no benefits for males aged over 78 as the corresponding baseline age-adjusted EQ-5D score for individuals with no history of CVD is 0.7748. Because the minimum model does not apply a constant detriment, the technique introduces a bias against older aged cohorts and the results from our threshold analyses demonstrate this can be quite substantial. We therefore recommend that the minimum model is not used to combine utility scores.

The authors of a recent publication propose a linear function to estimate HSUVs for combined health states which combines the three commonly used models. The weights for the function were obtained from a sample (n = 207) of men at the time of prostate biopsy. Although the authors found their weighted linear function outperformed the three individual models in terms of bias in the mean residuals and correlations of the residuals with the predicted HSUVs, these results are based on a baseline of perfect health [15].

Our results show that the multiplicative and additive models produce similar results both for the individual events and when applying the techniques in the economic model. Flanagan and colleagues found the multiplicative model was reasonably accurate in estimating both double and triple comorbidities after “purifying” the mean HUI3 scores to adjust for not having 26 chronic conditions [6]. Bond et al. concluded that the additive and multiplicative models produced very similar results, when using a baseline of perfect health [5]. However, the additive model applies a constant absolute detriment across all ages while the multiplicative model applies a constant proportional detriment. In real terms, this means that the additive model provides a greater absolute reduction in HSUVs than the multiplicative model and the magnitude of the detriment is constant across all ages irrespective of the number of comorbidities. The findings from Dale et al. and Fu et al., who both advocate the minimum model for combining HSUVs outside of an economic model, support the hypothesis that the detriment associated with several comorbidities may not equal the sum of the individual detriments.

Saarni reported that the mean number of comorbid chronic conditions increases from 1.1 for the age group 30–44 years to 4.0 for those aged 75 years and older [16]. It is possible that as the number of comorbidities increase, the detriment associated with an additional condition is smaller than that observed in an individual with just two comorbidities. If this hypothesis is correct, then the detriment associated with additional conditions would not be constant across all ages because of the increasing prevalence of comorbidities. In addition, health conditions can impact on the same health dimensions and it is reasonable to assume that an individual with two or more similar conditions will not necessarily have a reduction in HSUV that is equal to the sum of the reductions observed for each of the individual health conditions.

Although we found the additive and multiplicative models produced similar cost per QALY results, this finding may not generalize with other health conditions. In health conditions with comparatively small gains in QALYs, for example, when the intervention does not have an effect on mortality rates, the economic results are likely to be more sensitive to changes in the techniques used to combine HSUVs. Although additional research is required to support our hypothesis and findings, in the interim period, to facilitate comparison across results generated from models with multiple health states, we advocate the use of the multiplicative model for the reasons discussed above regarding the potential limitations associated with the additive model.

The health care literature and policy decision-makers such as NICE place a great deal of emphasis on both the methods used to obtain weights used in preference-based instruments and the particular preference-based instrument used to collect the HSUVs which are used to populate health states within economic models [1]. Evidence shows that the choice of instrument used to represent the HSUVs of a particular health condition can influence the results generated [17]. However, there is a great deal more to populating an economic model than the choice of instrument used to obtain the HSUVs and a consistent approach would improve comparability of results. We have used EQ-5D data in this article and additional research using alternative data such as the SF-6D is warranted.

Conclusion

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

Our results reinforce earlier recommendations and, until guidelines are in place, we would recommend that data from the general population are used as approximate baseline utility measures for individuals without the health condition under consideration if the actual data are not available. Although our findings demonstrate the additive and multiplicative models give similar results in CVD, additional research in other health conditions and datasets are required.

The underlying principle behind using the same preference-based instrument for all economic evaluations is to enable comparison across different interventions and health conditions. If this is to be realized, some consensus is needed on the most appropriate methods to populate the economic models. The methods used should be clearly described to inform policy decision-makers who are comparing results generated from different evaluations.

Acknowledgment

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References

We are grateful to Health Survey for England for allowing us access to the data used in this study. The research was based on the Health Survey for England 2003 and the Health Survey for England 2006, produced by the Joint Health Surveys Unit of Social and Community Planning and University College London, sponsored by the Department of Health, and supplied by the UK Data Archive. The data are Crown copyright.

Source of financial support: None.

References

  1. Top of page
  2. ABSTRACT
  3. Introduction
  4. Methods
  5. Analyses
  6. Discussion
  7. Conclusion
  8. Acknowledgment
  9. References
  • 1
    National Institute of Health and Clinical Excellence. Guide to the methods of technology appraisals. Available from: http://www.nice.org.uk/media/B52/A7/TAMethodsGuideUpdatedJune2008.pdf[Accessed December 1, 2009.
  • 2
    Fryback DG, Lawrence WG. Dollars may not buy as many QALYs as we think: A problem with defining quality of life adjustments. MDM 1997;17:27684.
  • 3
    Ward S, Lloyd Jones M, Pandor A, et al. A systematic review and economic evaluation of statins for the prevention of coronary events. Health Technol Assess 2007;11:1160, iii–iv.
  • 4
    Bansback N, Ara R, Ward S, et al. Statin therapy in rheumatoid arthritis: a cost-effectiveness and value-of-information analysis. Pharmacoeconomics 2009;27:2537.
  • 5
    Bond DE, Freedberg KA. Combining utility measurements exploring different approaches. Dis Manage Health Outcomes 2001;9:50716.
  • 6
    Flanagan W, McIntosh CN, Le Petit C, Berthelot JM. Deriving utility scores for co-morbid conditions: a test of the multiplicative model for combining individual condition scores. Population Health Metrics 2006;4:13.
  • 7
    Fu AZ, Kattan MW. Utilities should not be multiplied: evidence from the preference-based scores in the United States. Med Care 2008;46:98490.
  • 8
    Dale W, Basu A, Elstein A, Meltzer D. Predicting utility ratings for joint health states from single health states in prostate cancer: empirical testing of 3 alternative Theories. MDM 2008;28:10212.
  • 9
    Briggs AH. Handling uncertainty in cost-effectiveness models. Pharmacoeconomics 2000;17:479500.
  • 10
    Ara R, Tumur I, Pandor A, et al. Ezetimibe for the treatment of hypercholesterolaemia: a systematic review and economic evaluation. Health Technol Assess 2008;12:iii, xi–xiii.1–212.
  • 11
    Joint Health Surveys Unit of Social and Community Planning Research and University College London, Health Survey for England. 2003 [Computer File] (3rd ed.). Colchester, Essex: UK Data Archive, [distributor], 2005. SN: 5098.
  • 12
    Joint Health Surveys Unit of Social and Community Planning Research and University College London, Health Survey for England. 2006 [Computer File] (3rd ed.). Colchester, Essex: UK Data Archive, [distributor], 2008.
  • 13
    Dolan P, Gudex C, Kind P, Williams A. The time trade-off method: results from a general population study. Health Econ 1996;5:14154.
  • 14
    Ara R, Brazier J, Young T. Health related quality of life by age, gender or history of cardiovascular disease: results from the Health Survey for England 2003 and 2006. Discussion paper. Available from: http://www.sheffield.ac.uk/scharr/sections/heds/discussion.html[Accessed December 1, 2009].
  • 15
    Basu A, Dale W, Elstein A, Meltzer D. A linear index for predicting joint health states utilities from single health states utilities. Health Econ 2009;18:40319.
  • 16
    Saarni SI, Suvisaaria J, Sintonenb H, et al. The health-related quality-of-life impact of chronic conditions varied with age in general population. J Clinical Epidemiology 2007;60:128897.
  • 17
    Brazier J, Ratcliffe J, Salomon J, Tsuchiya A. Measuring and Valuing Health Benefits for Economic Evaluation. Oxford: Oxford University Press, 2007.