Nottingham Health Profile

Nottingham Health Profile in Outcome Measurement and Economic Evaluation

The Nottingham Health Profile records the health problems people say they have, sorted into six sections that are scored and reported separately. Hunt, McEwen, McKenna and colleagues developed it and described it in a series of papers from 1980, aiming at a standardised, simple measure of subjective health status that could be used in population surveys and in evaluating medical or social interventions. For a health economist the instrument raises a question that recurs with newer measures. Its statements carry weights from Thurstone paired comparisons, which are judgements of how serious each problem is, yet the scores are not utilities and cannot produce QALYs without a separate mapping step.

How the profile was built and tested

Hunt and McEwen described the practical development of a subjective health indicator in 1980, built in the form of a profile and reported to give valid measures of perceived health status that respondents found acceptable. The same year, Hunt and colleagues tested the profile in four groups of elderly people who differed in health status. It discriminated between groups that differed in diagnosed chronic illness, number of consultations in primary care and physiological fitness.

In a 1981 paper titled after the instrument, Hunt and colleagues compared people who consulted a general practice with people who did not, and every section of the profile showed a significant difference between the two groups. Profile scores also related to self-rated health and to absence from work. Women reported poorer subjective health than men in every section except pain and physical mobility.

In 1985 Hunt, McEwen and McKenna presented the profile to general practitioners as a standardised tool for surveying health problems in a population. They also proposed it for evaluating medical or social interventions and as an adjunct to the clinical interview. The proposal to use it in evaluating interventions is what brings the profile into economic work, where it appears as a trial outcome.

Sections, statements and Thurstone weights

The first part of the profile groups its statements into six sections, energy, pain, emotional reactions, sleep, social isolation and physical mobility. The profile also has a second part, and in a 1981 reliability study in patients with osteoarthrosis, test-retest correlations ranged from 0.77 to 0.85 for the first part and from 0.44 to 0.86 for the second, so the second part was less stable.

The statements in the six sections carry weights. McKenna, Hunt and McEwen derived the weights with Thurstone's method of paired comparisons, in which respondents judge which of two health problems is more serious and the pattern of choices is turned into scale values. Their 1981 paper used the sleep section, which contains five statements, as its worked case, described how the scale values were calculated, proposed a new way of deriving weights from them and found that interviewees accepted the comparison task.

A weighted section score of this kind adds the weights of the statements a respondent endorses.

$$S_k = \sum_{i \in k} w_i \, y_i$$

where $S_k$ is the score for section $k$, $w_i$ is the weight of statement $i$, $y_i$ equals 1 when the respondent endorses statement $i$ and 0 otherwise, and the sum runs over the statements in section $k$. A respondent who endorses nothing scores 0, which marks no reported problems in that section. The six section scores are reported side by side as a profile, with no weights to combine them.

Whether the weights change the answer

The weights look like a strength of the profile, since they reflect how serious people judge each problem to be. Evidence from translation work suggests they change little in practice.

Prieto, Alonso, Viladrich and Antó derived Thurstone weights for the Spanish version from 1,123 people and found that Spanish and English weights correlated at 0.72 to 0.93. When they compared unweighted, weighted and randomly weighted scores, the three strategies correlated extremely highly and discriminated between levels of health status with similar precision. Scaling assumptions held using unweighted scores, and the authors recommended unweighted scoring because the weights did not improve the instrument's psychometric properties.

For an economist the finding is a useful warning. Weighting statements within a section is a different task from valuing health states, and a weighting step that adds little to a profile's measurement properties does nothing to make its scores suitable for a QALY.

Why the NHP is a profile and not a preference-based measure

The NHP shows that collecting preferences is not enough to make a measure preference-based in the economic sense. A health state utility value needs three things the NHP does not provide. The values must describe whole health states, they must sit on a scale anchored at full health and dead, and they must let losses in one dimension be traded against gains in another.

The Thurstone weights do none of these. They scale the seriousness of individual statements, as the sleep-section example shows, and they are not values of whole health states. A section score of 0 means no problems were endorsed in that section and has no link to the value of being dead. No weighting combines the six sections, so two people with different profiles cannot be ranked without an outside judgement. The NHP is therefore a non-preference-based generic profile, and its section scores cannot be multiplied by time to give QALYs.

How health economists use NHP data

As a secondary outcome, NHP section scores show where a treatment changed perceived health alongside the QALY from a utility measure collected in the same trial. In a cost-consequence analysis the six sections can sit beside costs as separate outcomes, leaving the weighing of them to the decision maker. Older trials that collected the NHP and no utility measure are where mapping matters most, because they hold evidence that a modern cost-utility analysis cannot use directly.

Mapping NHP responses to EQ-5D

A mapping study estimates a statistical link between the profile and a preference-based measure collected in the same people. Vokó and colleagues used a cross-sectional sample of patients with diabetes in Hungary who completed both the NHP and the EQ-5D. They built the mapping function by stepwise backward least squares regression and judged fit by adjusted R squared, Akaike's information criterion and root mean square error. The best fitting model used all NHP statements as predictors, with an adjusted R squared of 0.68, and a reduced model kept 19 statements. The authors concluded that the mapping can estimate expected EQ-5D utility for a group of patients with diabetes, with limited use for predictions about individuals.

A distress index built from NHP statements

Some of the profile's statements have been reused as a single scale. Wann-Hansson, Klevsgård and Hagell describe the NHP distress index as a proposed generic, unidimensional 24-item measure of illness-related distress embedded in the NHP, and in 2008 they reported support for its validity in people with Parkinson's disease and with peripheral arterial disease. A single distress score is still a psychometric scale, so it is not a utility either.

Illustrative worked example of weighted and unweighted sleep scores

All numbers in this example are illustrative, and the weights are invented for teaching, not the published NHP weights. The sleep section has five statements, here given weights of 12, 16, 22, 20 and 30, which add to 100.

1. Weighted scores. Respondent A endorses statements 3 and 5, so the weighted score is 22 + 30 = 52. Respondent B endorses statements 1, 2 and 4, so the weighted score is 12 + 16 + 20 = 48.

2. Unweighted scores. Counting endorsed statements and scaling to 100 gives 2 ÷ 5 × 100 = 40 for A and 3 ÷ 5 × 100 = 60 for B.

3. What changes. With weights, A reports the more serious sleep problems. Without weights, B does. At the level of one person the weights can reverse a ranking. Prieto and colleagues compared scoring strategies across a whole sample rather than for single respondents, and found that weighted and unweighted scores correlated extremely highly.

4. Mapped utilities and cost per QALY. Suppose a trial of a sleep intervention reports a fall in the mean weighted sleep score of 12 points and no utility data. A mapping function, here assumed for illustration, predicts a utility gain of 0.05 lasting half a year, so the QALY gain is 0.05 × 0.5 = 0.025.

$$\text{ICER} = \frac{500}{0.025} = 20000$$

where 500 is the illustrative incremental cost in GBP, 0.025 is the mapped QALY gain and the result is the ICER in GBP per QALY.

5. Interpretation. The 12-point fall in the sleep section says the intervention reduced sleep problems but gives no value for that change. The ICER of GBP 20,000 per QALY depends entirely on the mapping function, which in this example is assumed. With a real mapping, such as the Hungarian diabetes model, the ratio would also carry the error of predicting EQ-5D from NHP statements in a population that may differ from the trial's.

How the NHP differs from neighbouring measures

The NHP is a generic instrument, since its statements are not tied to one disease, and a patient-reported outcome measure, since respondents report on themselves. Within that family it differs from its neighbours in what it asks about and where its scores cluster.

The SF-36 offers the clearest comparison, because the two have been tested side by side. Brazier and colleagues surveyed 1,980 patients aged 16 to 74 from two general practices in Sheffield and reported in 1992 that the SF-36 could detect low levels of ill health in patients who had scored 0 on the NHP. A score of 0 on the NHP therefore does not mean the absence of ill health, and the profile is less able to separate people near the healthy end. Brazier, Roberts and Deverill later estimated a preference-based measure, the SF-6D, from the SF-36, a route to QALYs that the NHP lacks without mapping. The SF-12 is a shorter generic profile from the same family.

The WHOQOL-BREF measures quality of life in a broader sense, including environment and social relationships, while the NHP statements concern health problems. A health profile in general is any set of separate dimension scores, and the NHP is one named instrument of that kind.

Limitations and common misreadings

Most errors with the NHP come from asking its scores to do more than they were built for. The weighting and the clustering of scores at the healthy end cause most of the misreadings.

• Section scores read as utilities. A section score reflects which problems a respondent endorses. It has no anchor at dead and cannot be multiplied by time to give QALYs.
• Summing the sections. Adding the six section scores into a total gives each section an equal weight that no valuation supports.
• Clustering at the healthy end. The Sheffield comparison found ill health among people scoring 0 on the NHP, the best possible score, which is the pattern behind a ceiling effect. In mildly ill populations the profile may show no change when health has changed.
• Overstating the weights. The Thurstone weights rank the seriousness of statements within a section. Evidence from the Spanish version suggests that unweighted scores perform about as well.
• Moving a mapping to a new population. The Hungarian model was estimated in patients with diabetes and its authors limited it to group-level estimates in that population.
• Reliability of the second part. The second part showed test-retest correlations as low as 0.44 in the osteoarthrosis study, so its items need care as an outcome.

Sources

• Brazier JE, Harper R, Jones NM, O'Cathain A, Thomas KJ, Usherwood T, Westlake L. Validating the SF-36 health survey questionnaire: new outcome measure for primary care. BMJ. 1992;305(6846):160-164. doi:10.1136/bmj.305.6846.160
• Hunt SM, McEwen J. The development of a subjective health indicator. Sociology of Health and Illness. 1980;2(3):231-246.
• Hunt SM, McKenna SP, McEwen J, Backett EM, Williams J, Papp E. A quantitative approach to perceived health status: a validation study. Journal of Epidemiology and Community Health. 1980;34(4):281-286. doi:10.1136/jech.34.4.281
• Hunt SM, McKenna SP, McEwen J, Williams J, Papp E. The Nottingham Health Profile: subjective health status and medical consultations. Social Science and Medicine Part A. 1981;15(3 Pt 1):221-229. doi:10.1016/0271-7123(81)90005-5
• Hunt SM, McKenna SP, Williams J. Reliability of a population survey tool for measuring perceived health problems: a study of patients with osteoarthrosis. Journal of Epidemiology and Community Health. 1981;35(4):297-300.
• Hunt SM, McEwen J, McKenna SP. Measuring health status: a new tool for clinicians and epidemiologists. Journal of the Royal College of General Practitioners. 1985;35(273):185-188.
• McKenna SP, Hunt SM, McEwen J. Weighting the seriousness of perceived health problems using Thurstone's method of paired comparisons. International Journal of Epidemiology. 1981;10(1):93-97. doi:10.1093/ije/10.1.93
• Prieto L, Alonso J, Viladrich MC, Antó JM. Scaling the Spanish version of the Nottingham Health Profile: evidence of limited value of item weights. Journal of Clinical Epidemiology. 1996;49(1):31-38. doi:10.1016/0895-4356(95)00064-X
• Vokó Z, Németh R, Nagyjánosi L, et al. Mapping the Nottingham Health Profile onto the preference-based EuroQol-5D instrument for patients with diabetes. Value in Health Regional Issues. 2014;4:31-36. doi:10.1016/j.vhri.2014.06.002
• Wann-Hansson C, Klevsgård R, Hagell P. Cross-diagnostic validity of the Nottingham Health Profile Index of Distress (NHPD). Health and Quality of Life Outcomes. 2008;6:47. doi:10.1186/1477-7525-6-47
• Brazier J, Roberts J, Deverill M. The estimation of a preference-based measure of health from the SF-36. Journal of Health Economics. 2002;21(2):271-292. doi:10.1016/S0167-6296(01)00130-8
• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford University Press; 2015.