The presence of exacerbations during the reference period was significantly associated with more physician visits (7.2 vs. 4.9, P < .005), specialist referrals (0.33 vs. 0.17, P=.048), and hospitalizations (0.19 vs. 0.015, P < .005) per year. Similar associations with degree of severity were observed; there were 7.9 versus 3.8 physician visits (P < .005), 0.3 versus 0.02 specialist referrals (P=.13), and 0.16 versus 0.02 hospitalizations (P =.05) per year in the highest versus lowest severity groups. Stratification by both exacerbation status and severity is shown in Table 2. Again, the number of resource units consumed increased with severity, and higher levels were reached in the presence of exacerbations. These tendencies were evident in all subgroups, albeit somewhat less unambiguously in the specialist referrals. Regression analysis demonstrated interaction between degree of severity and exacerbation status in their effect on the number of physician visits (P=.03 for a set of three dummy variables representing interaction) and hospitalizations (P =.04), but not the number of specialist referrals (P=.53).
Table 2. Resource use by degree of severity and exacerbation status in units per patient-year
|Group||N||Physician visits*||Specialist referrals*||No. of hospitalizations*|
|Total sample||422||5.8 ± 5.0|| 0.23 ± 0.80||0.083 ± 0.34|
|Exacerbations absent||263||4.9 ± 4.5|| 0.17 ± 0.68||0.015 ± 0.12|
| By degree of severity|
| Mild intermittent|| 30||3.7 ± 4.5||0.033 ± 0.18||0.033 ± 0.18|
| Mild persistent|| 79||4.1 ± 4.7|| 0.18 ± 0.53||0.013 ± 0.11|
| Moderate persistent|| 39||5.2 ± 4.0|| 0.16 ± 0.52||0.011 ± 0.10|
| Severe persistent|| 61||6.3 ± 4.7||0.25 ± 1.1||0.016 ± 0.13|
|Exacerbations present||159||7.2 ± 5.4|| 0.33 ± 0.96|| 0.19 ± 0.51|
| By degree of severity|
| Mild intermittent|| 14||4 ± 2.7|| 0 ± 0|| 0 ± 0|
| Mild persistent|| 31||4 ± 3.4|| 0.29 ± 0.78||0.097 ± 0.30|
| Moderate persistent|| 42||7.1 ± 5.3||0.45 ± 1.3|| 0.19 ± 0.40|
| Severe persistent|| 72||9.3 ± 5.7|| 0.35 ± 0.86|| 0.27 ± 0.65|
The presence of exacerbations during the reference period was associated with higher direct medical costs (CHF 3202 vs. CHF 1029, P=.0001), physician costs (CHF 269 vs. CHF 207, P < .00005), medication costs (CHF 724 vs. CHF 901, P =.056), and hospitalization costs (CHF 2031 vs. CHF 99, P < .00005) per year. There also was a steady increase with degree of severity. In the highest versus lowest severity groups, direct medical costs were CHF 3075 versus CHF 627, physician costs were CHF 284 versus CHF 109, medication costs were CHF 1122 versus CHF 336, and hospital costs were CHF 1669 versus CHF 182 (p < .005 in all cases). Stratification by exacerbation status as well as severity revealed further details (Table 3). In the patients with no exacerbations, there was a clear positive association of severity with direct medical costs and medication costs, but less of an association with physician costs and no association with hospitalization costs. In absolute terms, hospitalization costs were minimal here. In the patients who experienced exacerbations, a positive association with severity was seen in all cost categories. Absolute hospitalization costs were important here. In the moderate and severe patients with hospitalizations, mean hospitalization costs were CHF 10333 (SD 7018) and CHF 13875 (SD 11039), respectively.
Table 3. Costs by degree of severity and exacerbation status in CHF per patient-year
|Group||n||Direct medical costs*||Physician costs*||Medication costs*||Hospitalization costs*|
|Total sample||422||1848 ± 4134||230 ± 257|| 791 ± 746|| 827 ± 3887|
|Exacerbations absent||263||1029 ± 1274||207 ± 266|| 724 ± 654|| 99 ± 1010|
| By degree of severity|
| Mild intermittent|| 30|| 708 ± 1509|| 99 ± 101|| 342 ± 481|| 267 ± 1461|
| Mild persistent|| 79|| 804 ± 676||225 ± 362|| 541 ± 380|| 38 ± 338|
| Moderate persistent|| 93||1187 ± 1617||211 ± 217|| 826 ± 516|| 151 ± 1452|
| Severe persistent|| 61||1238 ± 1089||229 ± 230|| 992 ± 978|| 16 ± 128|
|Exacerbations present||159||3202 ± 6315||269 ± 237|| 901 ± 868||2031 ± 6019|
| By degree of severity|
| Mild intermittent|| 14|| 452 ± 476||130 ± 75|| 322 ± 433||0 ± 0|
| Mild persistent|| 31||1275 ± 3249||153 ± 134|| 380 ± 329|| 742 ± 3151|
| Moderate persistent|| 42||3089 ± 4947||297 ± 224|| 912 ± 789||1881 ± 4964|
| Severe persistent|| 72||4631 ± 8058||331 ± 272||1231 ± 969||3069 ± 7717|
Consequently, the relative proportions of cost categories differed greatly between those with and without exacerbations: In the latter, physician costs accounted for 20.1% of total direct medical costs, medication costs accounted for 70.4%, and hospitalization costs for 9.6%. In those with exacerbations, physician costs contributed 8.4% and medication costs contributed 28.1%, but hospitalization costs contributed 63.4%. The absolute levels of physician and medication costs in those with exacerbations were only slightly higher than in those without exacerbations. To a very large extent, hospitalization costs accounted for the differences observed.
Spearman's correlation coefficients of annual direct medical costs with the number of physician visits (.69), the number of specialist referrals (.14), and the number of hospitalizations (.44) were significant at the 5% level (P < .005 in all three cases). Spearman's correlation coefficients, however, with possible nonresource use influence factors were weak, except in the case of age (.25, P < .005), FEV1 (–.19, P=.008), and disease duration (.13, P=.033). Pearson's coefficients of the same variables with the logarithm of direct costs were almost identical. There appeared to be no relevant correlations between costs and BMI or FVC.
Mann–Whitney U or Kruskal–Wallis tests revealed significantly higher costs for the following influences: controller therapy compared to quick reliever therapy (P < .005); involvement of a pulmonologist in diagnosis or regular treatment (P < .005); nonemployment at the beginning of the reference period in patients aged 65 or younger (P=.005); and presence of asthma-related comorbidities (P=.029). The use of Student's t tests or ANOVA with the logarithm of direct costs led to the same results. Absences from work in the employed, German language region, and rural versus urban dwelling were associated with higher costs and had nonsignificant P values below .2, whereby these factors qualified as candidate predictors in multivariate analysis. Evaluation by insurance coverage did not reveal any existing associations.
There were no unexpected associations between possible influence factors on direct costs. Degree of severity and treatment type correlated significantly (P < .00005), but Spearman's correlation coefficient was only .33.
Multivariate Analysis of Direct Medical Costs
Multiple regression analysis on direct medical costs was based on the following possible influence factors derived from bivariate analysis: degree of severity, presence of exacerbations, quick reliever versus controller therapy, involvement of a pulmonologist (in diagnosis or regular treatment), age, duration of disease, FEV1, presence of asthma-related comorbidities, employment status, absences from work, language region, and urban or rural dwelling. Other potential influences such as height, weight, BMI, and FVC were also explored. Resource use variables (e.g., number of physician visits, number of hospital days) were not taken into account, because they were direct contributors to costs.
Models using the logarithm of direct medical costs identified degree of severity, exacerbation status, quick reliever versus controller therapy, age and age squared, presence of asthma-related comorbidities, and involvement of a pulmonologist (in diagnosis or regular treatment) as relevant and significant influence factors, allowing for an adjusted R2 value of .34. Influences of language region and urban versus rural dwelling were not confirmed.
A four-level ordinal variable, represented by three dummy variables, was introduced in the model to allow for interaction between medication-based degree of severity and exacerbation status. Thus, the effect of exacerbations could be described separately for each degree of severity. This resulted in a partial F test with P value of .0008 for the set of dummy variables and increased the adjusted R2 value to .36, showing a greater effect of exacerbations on costs in the more severe asthma patients.
Terms representing employment status in the patients aged 65 or younger, and absences from work in the employed, were not included in the final analysis albeit significant or near significant, because they altered the model only slightly (adjusted R2 =.38).
Age and age squared were centered to avoid a colinearity problem with these variables. After this procedure, variance inflation factors showed a mean of 3.91. The highest value was seen in the exacerbation status variable (VIF 10.60), with the dummy variables representing interaction between degree of severity and exacerbation status showing VIFs of 8.28, 5.26, and 4.40. Other criteria were clearly noncritical: there were no standardized regression coefficients larger than 1. After inclusion of the interaction terms, the parameter estimates and standard errors for the other variables changed very little, except of course for those terms which were bound to change because their meaning is different in the model with the interaction terms.
Details of the main model (n=420) are shown in Table 4. In larger models including other potential confounders, the degree of severity and exacerbation variables, and the respective interaction terms, had very similar coefficients. Relevant coefficient changes only occurred when FEV1 and body height or BMI were included. These models, though, had to rely on less than 100 observations.
Table 4. Multiple linear regression on the logarithm of direct medical costs (n=420)
|F(12, 407)=20.63; P > F=0.0000; R2=.3782; adjusted R2=.3599|
|Covariates||Coefficient||SE||t||P > |t|||95% CI|
|Degree of severity|
| Mild persistent|| 0.8218005*||0.1960516||4.19||<.0005||0.4364003 to 1.207201|
| Moderate persistent|| 1.012774*||0.1929955||5.25||<.0005||0.6333813 to 1.392166|
| Severe persistent|| 0.8962967*||0.2053643||4.36||<.0005||0.4925895 to 1.300004|
| Exacerbations present†|| 0.308257‡||0.2965676||1.04||.299||−0.2747384 to 0.8912525|
|Interaction variable, ordinal§|
| Level 1||−0.4850357||||0.3542102||−1.37||.172¶||−1.181346 to 0.2112742|
| Level 2|| 0.2323904||||0.341103||0.68||.496¶||−0.4381532 to 0.9029339|
| Level 3|| 0.5271099||||0.3369162||1.56||.118¶||−0.1352033 to 1.189423|
|Age (centered)|| 0.0054129||0.0023979||2.26||.025||0.000699 to 0.0101268|
|Age squared (centered)||−0.0002552||0.0001084||−2.36||.019||−0.0004683 to −0.0000422|
|Asthma-related comorbiditiy present †¶|| 0.37321||0.1267628||2.94||.003||0.1240184 to 0.6224016|
|Involvement of pulmonologist†|| 0.2758121||0.09196||3.00||.003||0.0950363 to 0.456588|
|Controller therapy†|| 0.6000914**||0.1160486||5.17||<.0005||0.3719619 to 0.8282209|
|Intercept|| 5.14959||0.1849907||27.84||<.0005||4.785934 to 5.513247|
Residual analysis (based on scatter plots of residuals vs. predicted values and age, box plots of residuals grouped by noncontinuous influence factors, and normality plots of residuals) gave satisfactory results. Exclusion of influential points identified by Cook's distance and the covariance ratio (resulting n=393) did not affect significance or greatly alter coefficients, but increased the R2 value to .43. Alternative versions of the model, for example, using the frequency of exacerbations rather than their presence or absence, gave very similar results.
Costs increased with age, but the effect was mild, nonlinear, and less pronounced in older age groups. Assuming constant other parameters, the presence of asthma-related comorbidities was associated with a 50% increase in direct medical asthma costs, controller therapy versus quick reliever therapy with an 80% increase, and involvement of a pulmonologist in diagnosis or treatment with a 30% increase. Greater degrees of severity and the presence of exacerbations during the reference period were also associated with higher costs (Table 5). Because the effect of these variables is greater than multiplicative, patients with a greater degree of asthma severity who experienced exacerbations were particularly expensive. Patients classed as severe persistent who experienced exacerbations cost more than five times as much as mild intermittent patients who did not.
Table 5. Effect of degree of severity and exacerbation status on direct medical costs according to the estimated regression model
|Degree of severity||Multiplication factor (95% CI)*|
|Exacerbations absent||Exacerbations present|
|Mild intermittent||1 (reference)||1.4 (0.8–2.4)|
|Mild persistent||2.3 (1.5–3.3)||1.9 (1.2–3.0)|
|Moderate persistent||2.8 (1.9–4.0)||4.7 (3.1–7.3)|
|Severe persistent||2.5 (1.6–3.7)||5.7 (3.8–8.4)|
Regression analysis on the logarithm of direct medical costs excluding medication costs resulted in a model comprising the same influence factors as described above, except for age and the interaction term between degree of severity and exacerbation status (n=420, R2=.22). These variables themselves remained highly significant. Degree of severity and treatment type accounted for 19% of the variance seen in the medication costs.