Overall survival distribution of the 170 patients with bullous pemphigoid initially treated with topical corticosteroids (testing sample).
Overall survival distributions of the 170 patients with bullous pemphigoid according to age (≤83 vs >83 years) (A) and Karnofsky score (≤40 vs >40) (B).
Overall survival distributions (observed and predicted) of the 170 patients with bullous pemphigoid according to joint levels of the 2 independent prognostic factors of age (≤83 vs >83 years) and Karnofsky score (≤40 vs >40). Data were obtained from the empirical cumulative hazard function estimated from the Cox proportional hazards regression model including age and Karnofsky score.
Overall survival distributions of the patients with bullous pemphigoid who were initially treated with oral corticosteroids at starting dosages of 0.5 mg/kg per day (n = 70) (A) or 1.0 mg/kg per day (n = 95) (B) according to joint level of the 2 independent prognostic factors of age (≤83 and >83 years) and Karnofsky score (≤40 and >40).
Joly P, Benichou J, Lok C, Hellot MF, Saiag P, Tancrede-Bohin E, Sassolas B, Labeille B, Doutre MS, Gorin I, Pauwels C, Chosidow O, Caux F, Estève E, Dutronc Y, Sigal M, Prost C, Maillard H, Guillaume JC, Roujeau JC. Prediction of Survival for Patients With Bullous PemphigoidA Prospective Study. Arch Dermatol. 2005;141(6):691-698. doi:10.1001/archderm.141.6.691
Copyright 2005 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2005
To identify the prognostic factors of bullous pemphigoid (BP).
Prospective study of patients with BP included in a randomized, controlled trial.
Twenty dermatology departments in France.
One hundred seventy patients with BP initially treated with a 40-g/d dosage of clobetasol propionate cream (testing sample) and 171 patients initially treated with oral corticosteroids at a dosage of 0.5 or of 1.0 mg/kg per day, depending on the extent of BP (validation samples).
Main Outcome Measures
The end point was overall survival during the first year after BP diagnosis. From the testing sample, associations of clinical and biological variables with overall survival were assessed using univariate and multivariate analyses. Selected predictors were included in a prognostic model. To verify that these predictors were not dependent on the treatment used, the model was then validated independently on the 2 series of BP patients treated with oral corticosteroids.
Median age of the BP patients included in the testing sample was 83 years. The 1-year Kaplan-Meier survival rate was 74%. From univariate analysis, the main deleterious predictors were demographic factors (ie, older age and female sex), associated medical conditions (ie, cardiac insufficiency, history of stroke, and dementia), and low Karnofsky score, which is a measure of the patient’s general condition. No factors directly related to BP, in particular extent of cutaneous lesions, were shown to be related to the patients’ prognosis. From multivariate analysis, only older age (P = .02) and low Karnofsky score (P<.001) appeared independently predictive of death. From the Cox model including these 2 predictors, the predicted 1-year survival rates were 90% (95% confidence interval [CI], 85%-96%) for patients 83 years or younger with Karnofsky score greater than 40, 79% (95% CI, 69%-90%) for patients older than 83 years with Karnofsky score greater than 40, 65% (95% CI, 50%-86%) for patients 83 years or younger with Karnofsky score of 40 or less, and 38% (95% CI, 26%-57%) for patients older than 83 years with Karnofsky score of 40 or less. Kaplan-Meier survival distributions of patients from the validation samples appeared clearly separated according to these 4 categories and were in close agreement with corresponding predicted 1-year survival rates obtained from the testing sample.
The prognosis of patients with BP is influenced by age and Karnofsky score. These predictors are easy to use and should facilitate the management of BP.
Bullous pemphigoid (BP) is the most frequent autoimmune blistering skin disease.1,2 It is most common in elderly patients.3- 6 Until recently, systemic corticosteroids have been considered the standard treatment of BP.3,7
The prognosis of BP is severe because the 1-year mortality rate has been reported to range from 25% to 40% in recent series.3,4,8- 12 Two recent retrospective studies3,12 have identified older age as independently predictive of death, but other independent prognostic factors identified by these 2 studies were in stark contrast. Indeed, the study by Roujeau et al3 found extensive (relative to moderate) disease form, poor general condition, and female sex to be independently predictive of death, whereas the study by Rzany et al12 identified high doses of oral corticosteroids at hospital discharge and a low serum albumin level as independent prognostic factors of death. It is likely that the retrospective nature of these 2 studies and the absence of control of treatment modalities were responsible for these discrepancies.
Given the large number of treatments proposed in the literature for patients with BP (hereafter referred to as BP patients)13- 21 and the high risk of severe adverse effects with treatments such as oral corticosteroids or immunosuppressive drugs in these patients, who are mostly elderly,13- 21 it appears critical to adapt treatment aggressiveness to individual patient prognosis.21- 23 To disentangle the conflicting findings of the 2 previous retrospective studies, we undertook a prospective prognostic study of BP using a large series of patients treated in a homogeneous manner. Patients in our series were included in a recently reported randomized clinical trial that compared oral and topical corticosteroids in the treatment of BP patients.11 This randomized trial demonstrated that high-potency topical corticosteroids were safer and more effective than high doses of oral corticosteroids for controlling extensive BP. Half of the patients in this trial were treated with topical corticosteroids. Using data from these patients, prognostic factors of overall survival were identified and a prognostic regression model was developed. This model was then validated independently on 2 series of patients who received various dosages of oral corticosteroids.
This prospective multicenter study was conducted in 20 departments of dermatology in France. More details can be found in Joly et al.11 Briefly, consecutive patients with incident BP were recruited if the following inclusion criteria were met: (1) clinical features suggestive of BP6; (2) the typical histological pattern of BP, including a subepidermal blister on a skin biopsy specimen24; and (3) linear deposits of IgG and/or C3 along the basement membrane zone as shown on results of direct immunofluorescence.25,26 Patients were randomly assigned to receive topical or oral corticosteroids. The study was approved by the ethics committee of Seine Maritime, and written informed consent was obtained for each patient.
Prognostic factors of overall survival were determined from the data of patients randomized to receive topical corticosteroids (testing sample). These patients were treated initially with a 40-g/d dosage of clobetasol propionate cream that was applied twice daily on the entire body surface until 15 days after disease control. Disease control was defined as the absence of new bullae for 3 consecutive days. Dosages of clobetasol propionate cream were then gradually tapered to 20 g/d for 1 month, 10 g/d for 2 months, 10 g every other day for 4 months, and finally 10 g twice a week for 4 months. Relapse was defined as the occurrence of at least 3 new bullae daily for 3 consecutive days during treatment. Patients who experienced relapse during the period when treatment dosages were being reduced had their dosages increased to the previous level that had permitted disease control. Treatment was stopped after 12 months.
External validation of the prognostic factors and corresponding prognostic model was carried out from the data of 2 series of patients randomized to receive oral corticosteroids (validation samples). Patients with moderate BP (ie, ≤10 new bullae daily) received oral prednisone (Cortancyl; Roussel Laboratories, Paris, France) at the initial daily dosage of 0.5 mg/kg of body weight, whereas those with extensive BP (ie, >10 new bullae daily) received twice that dosage initially (ie, 1 mg/kg of body weight per day). The initial dosage was maintained for 15 days after disease control had been attained. Thereafter, it was reduced by 15% every 3 weeks. As for patients in the testing sample, the dosage was increased to the previous level that had permitted disease control if a relapse occurred, and treatment was stopped after 12 months. In both the testing and validation samples, other therapy that might affect BP activity was avoided throughout the 12-month study period, and investigators stopped treatment if a life-threatening adverse effect occurred.
At baseline, each patient underwent complete physical examination. The daily number of new bullae during the 3 days before the initiation of therapy was noted by a nurse. The following clinical findings were recorded: delay since onset of initial cutaneous lesions, type and localization of cutaneous lesions, pruritus, Karnofsky score, personal history of cardiovascular disease (eg, high blood pressure, cardiac insufficiency, angina pectoris, ischemia, cardiac arrhythmia, arteriosclerosis of the leg, deep vein thrombosis, or pulmonary embolism), neurologic disorder (eg, stroke, dementia, Parkinson disease, or other), psychiatric disorder, hepatic and/or digestive disease (eg, diverticulosis, cirrhosis, gastric ulcer, or cholelithiasis), chronic lung condition, diabetes mellitus, malignancy, endocrine disorder, and urologic disease. The Karnofsky score is a measure of the patient’s general condition and degree of autonomy on a scale ranging from 0 to 100, with higher scores indicating better condition and greater autonomy.27 Blood eosinophil counts and titers of serum anti–membrane zone antibodies were determined by indirect immunofluorescence on normal human skin and recorded.
Follow-up visits were planned on days 7, 14, 21, 30, 90, 180, and 360. The date of any relapse that occurred and the date and cause of death in any patients who had died were recorded. Because of the high 1-year mortality rate reported in recent series, follow-up was planned to last 12 months only.
The end point of this prognostic study was overall survival during the first year after the diagnosis of BP. To select determinants of overall survival, the testing sample was used. Associations between baseline variables and overall survival were studied using univariate and multivariate analyses. In univariate analysis, the prognostic significance on overall survival was assessed using the log-rank test for categorical variables and the Cox proportional hazards regression model for continuous variables. Relative risks (RRs) and corresponding 95% confidence intervals (CIs) were estimated using the Cox proportional hazards regression model for categorical and continuous variables. The RRs per unit were estimated for continuous variables. Variables that appeared separately associated with overall survival at the P<.15 level were further retained for multivariate analysis. Stepwise Cox regression was used to select prognostic variables independently associated with overall survival at the P<.05 level. Adjusted RRs and corresponding 95% CIs were estimated for these variables. Observed distributions of overall survival according to levels of these variables (alone or in combination) were estimated using the Kaplan-Meier method.
Predicted distributions of survival were obtained from the Cox model–based estimates of empirical cumulative hazard functions.28 The multivariate coefficient of determination (R2) of Xu and O’Quigley was used to assess the proportion of variation explained by the variables included in these models.29 We used SAS statistical software (version 8.02; SAS Institute Inc, Cary, NC) for all other analyses. Unless otherwise indicated, continuous variables are expressed as mean ± SD.
One hundred seventy patients (69 men and 101 women) were randomized to receive topical corticosteroids and thus were included in the testing sample of this prognostic study. Their mean age was 81 ± 10 years (median age, 83 years). Their mean Karnofsky score was 64 ± 25 (median score, 60). The mean number of daily new blisters was 29 ± 46. One hundred eleven (73%) of 153 patients were determined by indirect immunofluorescence to have circulating anti–basement membrane zone antibodies, and the mean number of blood eosinophils was 1480 ± 1450 cells/μL.
Control of BP was achieved in all but 1 patient (169/170 [99%]) by day 21. Sixty-one patients (36%) experienced relapse during the study after a mean of 185 ± 122 days.
During the 1-year follow-up, 45 patients (26%) died. The primary cause of death was determined in 34 cases. The main causes were sepsis (in 10 patients, including 8 with pneumonia), cardiovascular disease (in 9), and stroke (in 5). The 1-year Kaplan-Meier overall survival rate was 73% (95% CI, 66%-80%) (Figure 1).
Among the surviving patients, the mean follow-up time was 360 ± 105 days (median, 365 days; 25th and 75th percentiles, 360 and 365 days, respectively).
From univariate analysis, older age was strongly associated with shorter survival, whether age was dichotomized (eg, ≤83 vs >83 years) or coded as a continuous variable (Table 1). For example, the RR of death was estimated at 3.1 (95% CI, 1.6-5.9; P<.001) and the 1-year Kaplan-Meier overall survival was estimated at 61% for patients older than 83 years (85% for patients ≤83 years) (Figure 2A). We retained the dichotomous coding (≤83 and >83 years) for further analysis because the median age in this series was 83 years. Female sex was likewise associated with poorer survival. Poor general condition, as reflected by a low Karnofsky score, was strongly associated with poorer survival for various codings. For instance, the RR of death was estimated at 4.9 (95% CI, 2.7-9.2; P<.001) and the 1-year Kaplan-Meier survival rate was estimated at 48% for values of 40 or less (86% for values >40) (Figure 2B). This threshold was retained for further analysis because it allowed us to separate patients who remained at least partly autonomous (Karnofsky score, >40) from those who were bedridden (Karnofsky score, ≤40). Moreover, several medical conditions (eg, dementia, cardiac insufficiency, and personal history of stroke) appeared to be associated with poorer survival (Table 1). No variable reflecting extent or severity of BP, blood eosinophilia, or serum anti–basement membrane zone antibodies appeared to be associated with overall survival (Table 2).
From multivariate analysis, only age and the Karnofsky score appeared independently predictive of overall survival. Patients older than 83 years had an elevated risk of death relative to younger patients (RR, 2.3; 95% CI, 1.6-4.4; P = .02). Bedridden patients (Karnofsky score, ≤40) had a 4-fold increase in their risk of death (RR, 4.1; 95% CI, 2.2-7.7; P<.001) compared with more autonomous patients (Karnofsky score, >40). As no significant interaction was found between these 2 variables (P = .57), patients older than 83 years and bedridden were 9.3 times (95% CI, 4.1-20.9) more likely to die than younger, more autonomous patients. Female sex was not found to be independently predictive of survival because of its marked association with older age. Similarly, dementia, cardiac insufficiency, and history of stroke were not found to be independently predictive of survival because of their strong associations with patients’ general condition (for stroke and dementia) or age (for cardiac insufficiency).
From the Cox proportional hazards regression model including the 2 independent prognostic factors of age and the Karnofsky score, predicted 1-year survival rates of BP patients were 90% (95% CI, 85%-96%) for patients 83 years and younger with a Karnofsky score of greater than 40 (prognostic category 1), 79% (95% CI, 69%-90%) for patients older than 83 years with a Karnofsky score of greater than 40 (prognostic category 2), 65% (95% CI, 50%-86%) for patients 83 years or younger with a Karnofsky score of 40 or less (prognostic category 3), and 38% (95% CI, 26%-57%) for patients older than 83 years with a Karnofsky score of 40 or less (prognostic category 4), in close agreement with Kaplan-Meier 1-year survival estimates for these 4 categories (Figure 3). Moreover, R2 was equal to 0.48, indicating that this model including age and the Karnofsky score offered a fairly good prediction of patients’ overall survival.
The findings in the testing sample were independently validated in patients from the 2 validation samples. Seventy-six patients (28 men and 48 women) with moderate BP (ie, ≤10 new bullae daily) were randomized to receive oral corticosteroids (prednisone) at the initial dosage of 0.5 mg/kg per day and constituted the first validation sample. Their mean age was 81 ± 10 years (median age, 83 years) and their mean Karnofsky score was 66 ± 24 (median score, 60). Their mean number of daily new blisters was 4 ± 3 at baseline. The 1-year Kaplan-Meier overall survival rate of these patients was 70% (95% CI, 59%-80%). Kaplan-Meier survival distributions appeared clearly separated according to the 4 prognostic categories defined herein, and the ordering of these categories was essentially maintained. Kaplan-Meier 1-year survival rates of 90%, 70%, 67% and 35% for prognostic categories 1 to 4, respectively (Figure 4A), were in very close agreement with corresponding predicted 1-year survival rates obtained from the testing sample and well within their range of error. Moreover, R2 was equal to 0.44, indicating fairly good prediction of overall survival from the testing sample model. Moreover, upon fitting the same Cox model as in the testing sample, RR estimates were 3.0 for age greater than 83 years and 3.4 for a Karnofsky score of 40 or less, again in fairly close agreement with values obtained from the testing sample and well within their range of error.
The second validation sample included 95 patients (32 men and 63 women) with extensive BP (ie, >10 new bullae daily) randomized to receive oral prednisone at a dosage of 1 mg/kg per day. Their mean age was 81 ± 9 years (median age, 82 years), and their mean Karnofsky score was 63 ± 24 (median score, 60). Their mean number of daily new blisters was 64 ± 74 at baseline. Overall survival was worse than that of patients treated topically, with a 1-year Kaplan-Meier overall survival rate of 59% (95% CI, 49%-69%). Again, patients’ overall survival was strongly influenced by their age and Karnofsky score, with perhaps a weaker influence of the latter than in the testing sample and first validation sample. The Kaplan-Meier 1-year survival rates were 84%, 44%, 53%, and 33% for prognostic categories 1 through 4, respectively (Figure 4B).
The main result of this study is that old age and poor general condition are the 2 major detrimental prognostic factors of BP. Indeed, patients older than 83 years and with a Karnofsky score of 40 or less had a more than 9-fold increase in their risk of dying during the first year of treatment relative to younger (≤83 years), more autonomous (Karnofsky score, >40) patients, and their predicted 1-year survival rate was only 38%, whereas that of younger, more autonomous patients was 90%. These results could be validated on 2 independent series of BP patients treated with regimens of medium- or high-dose oral corticosteroids. Survival predictions appeared particularly accurate in patients treated with medium-dose oral corticosteroids.
Our results are consistent with those of Roujeau et al,3 who also found older age and poor general condition to be associated with decreased overall survival. However, unlike those authors, we did not find female sex to be an independent detrimental prognostic factor of overall survival. Indeed, although female sex was separately predictive of decreased overall survival in the univariate analysis (RR, 2.2; 95% CI, 1.2-4.3), it was not independently predictive of overall survival in the multivariate analysis (RR, 1.2; 95% CI, 0.6-2.5) because of its strong associations with older age. Our results are also consistent with those of Rzany et al12 regarding the detrimental prognostic role of older age.
No factor reflecting BP extent or severity was shown to be related to prognosis. Although this series consisted of patients with various extents of cutaneous lesions (ranging from 1-300 new bullae daily), no relationship was found between patients’ overall survival and their daily number of new bullae, whether it was coded as a continuous variable or dichotomized (≤10 or >10 new bullae daily). This result is in stark contrast with that of Roujeau et al,3 who found that the overall survival of patients with moderate BP was significantly longer than that of patients with extensive BP. This discrepancy may be explained by the retrospective nature of the study by Roujeau et al3 and the confounding effect of treatment. Indeed, in their study, many patients with moderate BP were treated with topical corticosteroids, whereas all patients with extensive BP were treated with high doses of oral corticosteroids, sometimes in association with immunosuppressive drugs. In contrast, our results, which are based on an accurate prospective assessment of BP extent, are not vulnerable to confounding by treatment and are in accordance with results from our group’s previous controlled study of azathioprine sodium and plasmapheresis in addition to corticosteroids, in which no relationship was found between the initial extent of BP and overall survival.30 Unfortunately, the daily number of new bullae was not available in the prognostic study by Rzany et al.12
This prognostic study has 2 major strengths. First, to our knowledge, it is the first extensive prognostic study of BP based on a prospective design. Second, the confounding effect of treatment can be ruled out because all patients in the testing sample received identical initial treatment and dosages were modified according to a standardized protocol. As for potential limitations, a priori sample size computations were performed for the randomized clinical trial11 on which this study was based but not for this prognostic study. However, an a posteriori computation based on the number of deaths observed in the testing sample was reassuring. Indeed, whereas this prognostic study had limited statistical power for detecting prognostic factors of moderate impact (eg, 60% power for a 2-sided log-rank test at the P<.05 level relative to balanced dichotomous predictors with an RR of 2.0 for the detrimental prognostic category), it had high power for detecting stronger risk factors (eg, 80% and 97% power for a 2-sided log-rank test at the P<.05 level relative to balanced dichotomous predictors, with respective RR values of 2.5 and 4.0 for the detrimental prognostic category).
Overall, the results from this study indicate that BP prognosis is influenced by age and general condition but not by disease activity. These predictors are easy to use and should facilitate the management of BP. In particular, these results should prompt clinicians to be cautious in the treatment of older BP patients who are in poor general condition.31,32 In our opinion, these patients should be treated with topical corticosteroids or with low doses of prednisone.
Correspondence: Pascal Joly, MD, PhD, Clinique Dermatologique, Hôpital Charles Nicolle, 1, rue de Germont, 76031 Rouen CEDEX, France (Pascal.Joly@chu-rouen.fr).
Accepted for Publication: December 21, 2004.
Group Members: The following investigators also participated in this study: Catherine Picard, MD, Paris, France; Brigitte Dreno, MD, PhD, Nantes, France; Emmanuel Delaporte, MD, PhD, Lille, France; Loïc Vaillant, MD, PhD, Tours, France; Michel D’Incan, MD, PhD, Clermont Ferrand, France; Patrice Plantin, MD, Quimper, France; Christophe Bedane, MD, PhD, Limoges, France; and Philippe Bernard, MD, PhD, Reimes, France.
Funding/Support: This study was supported by research grants from Rouen University Hospital, Rouen, France, and the French Society of Dermatology, Paris, France.
Acknowledgment: The software used to estimate the multivariate coefficient of determination (R2) was kindly provided by John O’Quigley, PhD, Department of Biostatistics, Institut Curie, Paris.
Financial Disclosure: None.