Analysis of the cues used by patients when making assessments of their asthma severity

Participants

Patients with a clinical diagnosis of asthma made by a consultant chest physician were recruited from two outpatient clinics. All patients attending the clinic on the day the author (E. Barley) was present were approached. A total of 40 patients (female = 25) aged 27–82 yrs (mean 52 yrs) agreed to participate. The mean number of years since diagnosis was 25±18 yrs or 53±37% of their life. The mean post-bronchodilator forced expiratory volume in one second (FEV₁) was 67.4±25.7% predicted. All were prescribed a β₂-agonist and an inhaled steroid. Nine were also prescribed other asthma medication, such as a xanthine, an anticholinergic or a cromone and six were currently taking oral steroids. Seventeen patients had never smoked, five were current and 18 were ex-smokers. All had been given asthma education and self-management training. Patients with other inter-current disease or a history of psychiatric illness were excluded. Patients were divided into two groups according to their medication regimen; simple: β₂-agonist plus an inhaled steroid; complex: β₂-agonist and inhaled steroid plus xanthine, anticholinergic, cromone or oral steroid. Ethical approval for the study was provided by Wandsworth Health Authority Local Research Ethics Committee, London, UK. Written consent was obtained from all patients.

Judgement analysis

In JA, participants make judgments about a set of hypothetical cases, which are described in terms of the factors or cues under study. The cues tested were the four main asthma health status domains: symptoms, activity, emotions and social. Each hypothetical case was characterised in terms of the impact of asthma on each of these domains. Symptoms were described in terms of wheeze, cough and breathlessness; emotions were defined in terms of anxiety, as this is the principle negative emotion associated with asthma 16; disturbance to daily activity and to social life were undefined. Each cue was presented at one of four levels of severity. Cues and levels are shown in Appendix 1. Four cues with four levels would produce 256 cases (4×4×4×4). Specialist JA software Policy PC 17 was used to randomly generate 50 cases. This is the number that can be comfortably judged by participants and is sufficient for robust analysis 18. Unknown to the patients, ten replicated cases were included to allow a test of within-patient reliability. An example of a case, in the form presented to the patients, is shown in Appendix 2. Patients made judgments about overall asthma severity of each of the 60 cases using a visual analogue scale (VAS).

Other measurements

Study design

Patients were interviewed at home. Age, height, self-reported number of years with a diagnosis of asthma (disease duration), medication regimen and smoking history were recorded. Spirometry was performed. The patients also completed the judgment task, the current asthma severity rating, the SGRQ and the HAD. The order of presentation was randomised using a Latin square. Interviews lasted ∼1 h.

Analysis

Within-patient reliability was assessed using the correlation between judgments of the 10 pairs of replicated cases. In JA studies, a Pearson's r of 0.6 is considered to reflect an acceptable level of reliability 18, so patients were categorised into reliable (r≥0.6) and less reliable (r<0.6) groups.

Judgment data were analysed using the multiple regression method developed by Hammond and Adelman 21. For each patient, multiple regression, with the rating of asthma severity as the dependent variable and the cue values as the independent variables, was determined. The coefficient R² represents the proportion of the variance in the judgments that can be accounted for by the derived weights 22. It also reflects the consistency of each patient's judgments; R² = 1 would indicate that the patient always gave each cue the same importance when rating overall severity, and that the overall judgment was completely accounted for in terms of the cues and weights attributed to them 23.

The JA-derived weights for the cues were used as measurements of how much each had contributed to the judgment. Relative weights were reported. These add up to 100 and may be interpreted as if the patient had divided up 100 points among the cues in proportion to the amount of emphasis they placed on the cue when making their overall judgments about asthma severity 24. These weights are not independent of each other for the purposes of statistical analysis because the size of one affects the size of the other three.

The influence of demographic and asthma-related factors on the JA-derived policies was tested by performing multinomial logistic regression, with the rank-ordered weights for each cue as the dependent variable. Rank ordering was necessary to overcome the problem of cue weight interdependence. A separate regression was calculated for each variable because some of the independent variables were correlated. The significance of a covariate was tested using a logistic likelihood ratio test (Chi-squared test).

To identify groups of patients whose cue weighting policies were most similar, a cluster analysis was performed. Only those patients who were reliable in the judgment task were analysed. Cue weights were converted to standard (z) scores before clustering them; this ensured that cue weights having the greatest variability across patients did not dominate the clustering process 24. A hierarchical cluster analysis using Ward's minimum variance method 25 was carried out. Patients within different clusters were compared for differences on the asthma-related variables. Results are presented as mean±sd, unless otherwise indicated.

Data collection

JA data were available for all patients, but in one patient no regression weights could be produced. On examination of this patient's data, it was apparent that they had marked the centre point of each VAS irrespective of cue level; this patient was excluded from further analysis. Questionnaire and rating data were available for all patients, with no missing answers. Post-bronchodilator FEV₁ data were available for 35 patients, as five were unwilling to take medication for the purposes of the study.

Patient details

Mean SGRQ total score was 40.6±17.7. Mean HAD score was 7±5 for anxiety and 4±3 for depression. Median global asthma severity score was 3.5 (range 0.2–10). Mean FEV₁ was 67.4±25.7% pred.

Reliability

Correlations between replicated cases ranged from r = 0.02 to r = 0.94 (median r = 0.60). In total, 21 patients were found to be reliable (i.e. r≥0.6). In these patients, correlations ranged 0.60–0.94 (median = 0.82). Internal consistency measured as R² ranged 0–0.94 (median R² = 0.62) for the entire sample, and 0.18–0.94 (median R² = 0.77) for the subgroup of 21 reliable patients. The estimates for r and R² were correlated (r = 0.69, p<0.0001). The scores for the asthma-related variables of reliable and less reliable patients are shown in table 1⇓. The less reliable patients were older, reported worse health and were prescribed more medication.

Judgment policies

The importance placed on each cue varied widely between patients (table 2⇓). Findings were similar for reliable and less reliable patients. Overall, the mean weight was greatest for the symptoms cue. The cue with the lowest mean weight was emotions. The symptoms cue was ranked most often as the most important cue (64% of all patients); no patient ranked it as least important. The cue most often ranked least important was emotions (41% of all patients).

Influence of asthma-related variables on cue weights

These analyses were performed using the 21 reliable patients only. A rank order relationship between each cue and a range of demographic and disease factors was tested for. Logistic regression showed that for the symptoms, activity and social cues there was a significant association between cue rank and disease duration (table 3⇓), but there was little evidence of ordering in this association. The strongest association was between the symptoms cue and disease duration. Patients who placed symptoms as their first rank cue had had the disease for 77% of their life, compared with those for whom it was second in importance, who had had asthma for 27% of their life. There was also a weak association between the emotions cue and HAD score (p = 0.05). Rank ordering of cues was not associated with lung function (fig. 1⇓; p>0.05).

Cluster analysis

Analysis of the 21 reliable patients indicated that they could be grouped according to two broad policy types (fig. 2⇓). Further analysis showed that patients in the first cluster placed considerably more weight on symptoms than on any other cue, with the other cues weighted similarly. Patients in the second cluster tended to weight all cues similarly. Compared with patients in the first cluster, patients in the second cluster gave significantly more weight to all the cues, except symptoms (fig. 3⇓).

The policies of patients in the first cluster were more similar than the policies of those in the second cluster (fig. 2⇓). Patients in the first cluster were slightly more reliable in their performance of the judgment task (Pearson's r between replicated cases median = 0.90, interquartile range (IQR) 0.08) than the patients in the second cluster (Pearson's r between replicated cases median = 0.78, IQR 0.19, Mann-Whitney U-test p = 0.0012). More of the variance in judgments was explained by the policies of the first cluster of patients (median R² = 0.88, IQR = 0.07) than those of the second cluster (median R² = 0.64, IQR = 0.25, Mann-Whitney U-test, p = 0.003). The two clusters did not differ on any demographic or disease-related variable.