Introduction
Bullous pemphigoid (BP) is a chronic debilitating autoimmune blistering disease which has a high mortality and morbidity [
28]. BP mainly affects the elderly people, causing symptoms, such as widespread tense blistering and itching. Previous studies reported an incidence of BP ranging from 2.5 to 42.8 cases per million per year [
2,
11,
14,
23]. BP has been reported to be significantly associated with mortality, with morality rates ranging from 6 to 41% within the first year after diagnosis [
1,
3‐
6,
12,
15,
22,
27,
29,
30,
32]. BP patients also have increased mortality compared with their age and sex-matched controls in the general population [
2‐
4,
12,
14].
However, relatively few studies have investigated prognostic factors of BP mortality, with considerably various results. Old age has been shown to be related to poor outcome in BP in several studies [
6,
12,
15,
27], but there are conflicting results on whether old age is an independent risk factor for mortality in BP [
3,
21,
22]. In addition, poor general condition and low karnofsky scale [
26] have often been associated with increased mortality [
1,
15,
21,
29]. Interestingly, factor directly related to BP, such as the severity and extent of skin lesion, does not often predict survival [
3,
12,
21,
22].
In addition, BP patients have complex comorbidity profiles, most notably neurological disorders as well as autoimmune, infectious and cardiovascular disorders [
28]. Various studies have confirmed a strong association between BP and neurological diseases, particularly dementia, stroke, Parkinson disease, multiple sclerosis, and epilepsy [
20]. BP is associated with development of autoantibodies directed against the hemidesmosomal BP autoantigens BPAG1 and BPAG2. BPAG1 has two isoforms: BPAG1-e is mainly expressed in skin, while BPAG1-a is expressed in central nervous system [
20]. It is commonly suggested that an immunological cross-reaction between these isoforms could be the pathogenic reason. However, the prognostic value of co-existing neurological disorders as well as other comorbidities for BP mortality is unclear.
In addition, the sample sizes of these studies are often relatively small, leading to limited statistical power to detect true associations. In addition, there is still no systematic and quantitative assessment of published findings on this topic. We, therefore, conducted a meta-analysis to quantitatively assess the association between several potential prognostic factors and risk of mortality in bullous pemphigoid.
Materials and methods
This meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement protocol [
25].
Search strategy and study selection
We performed an exhaustive search on Pubmed (from 1978 to November 2016), Embase (from 1974 to November 2016) and Cochrane Library (from 1994 to November 2016) to identify articles that studied the prognostic factors for BP mortality. There was no limit on human race, geographic region, language, or publication type. The major terms used included: “bullous pemphigoid”, “mortality”, “fatality”, “survival”, “death”, “prognosis”, and “prognostic”. The following selection criteria were performed on Pubmed as an example:
1.
bullous pemphigoid [Title] or bullous pemphigoid [Mesh Terms];
2.
mortality or fatality or survival or death or prognosis or prognostic [All Fields];
Studies eligible for inclusion in this meta-analysis met the following criteria: (a) cohort study design; (b) measure potential prognostic factors for BP mortality; (c) provide sufficient data for calculating the effect size with 95% confidence intervals (CIs); and (d) when the same author’s data obtained from the same or overlapping patients in more than one publication, only the most recent report or the most complete one was selected in the analysis. We excluded case reports, reviews, meta-analyses, and letters. Studies investigating remission or relapse but not mortality of BP were excluded. Childhood bullous pemphigoid was not considered in the present study. Additional studies were identified by a manual search of the references of original studies. The literature search and inclusion of eligible articles was performed independently by two authors (Liu YD and Zhao WL) and disagreements were resolved at each step by consensus.
All data were extracted independently and crosschecked by two authors (Liu YD and Zhao WL). Data retrieved from the studies included authors, year of publication, patient source, number of patients included in the study, study type, follow-up time, and statistical results regarding the influence of prognostic factors on the risk of mortality. If both 1 year mortality and overall mortality were reported in a study, 1 year mortality was extracted for analysis.
Quality assessment
Quality assessment of each study was performed independently by the two authors (Liu YD and Zhao WL), using the Newcastle–Ottawa Scale (NOS) (
http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp), which graded studies according to the quality of selection, comparability, and outcome of study participants. Studies that achieved seven or more stars on NOS were considered to be of high quality, four to six stars were medium quality, and fewer than four stars were poor quality [
24]. Discrepancies were addressed by re-evaluation of the original article and discussion with a third author (Wang YH).
Statistical methods
For the quantitative aggregation of the survival results, we reported summary measures of effect size using a pooled hazard ratio (HR) with 95% confidence interval. If available, we extracted the risk estimates that underwent multivariate adjustment. For studies in which the HR corresponding to the 95% CI was not given directly, published data from original papers were used to calculate the HR according to the methods described previously [
31]. In a few situations where there was not sufficient information for HR estimation, we calculated the relative risk (RR) with 95% CI as a crude substitution of HR. Study heterogeneity was assessed using the Cochran Q test with
P value for significance set at 0.10 and the
I
2 statistic with values of 25, 50 and 75% considered as low, moderate, and high heterogeneity, respectively [
13]. To estimate the pooled HR, random-effects model of DerSimonian and Laird was used [
7]. Publication bias was evaluated with funnel plot and Egger’s test (
P < 0.05 was considered statistically significant) [
9]. The statistical analysis was performed by the STATA 14.0 software (StataCorp, College Station, TX, USA).
Discussion
In this meta-analysis of 14 studies that included 2499 subjects, we found that older age, presence of circulating antibodies regardless of detection method, co-existing dementia, and stroke were associated with an increased risk of BP mortality. No significant association was observed for gender, disease extent, mucosal involvement, or IIF result and survival.
In consistence with most studies included in the current analysis, we found no survival difference based on sex. However, age at diagnosis was associated with increased risk of death, with advanced age being a poor prognostic factor. Stratification of studies based on age cut-off value demonstrated that the effect size was more significant when the cutoff was lower, suggesting that the truly younger patients may have a better prognosis. Discrepancies in the age cut-off values, patient baseline characteristics, and treatment approaches between studies might contribute to the high level of heterogeneity. The pooled result did not remain comparable after “trim and fill” adjustment, suggesting that publication bias might have led to an overestimation of the effect of age on survival. However, the asymmetry in funnel plot may also simply be the play of chance, since only ten studies were included, or true heterogeneity existed because cut-off age varied greatly. Of note, the two outlier investigations by Roujeau et al. [
29] and Cortés et al. [
5] were small studies with less than 100 participants. Nevertheless, the above finding indicates that younger age at the diagnosis of BP does not necessarily denote a lower risk of mortality, and physicians should be more cautious about patients’ general health condition and comorbidity profile.
As reported previously, more extensive form of BP did not significantly increase the risk on mortality. Although the extent and distribution of disease are a variable susceptible to high variability between centers and may introduce information bias, especially in retrospective studies, heterogeneity between the included studies was low. The presence of mucosal lesion was also not predictive of poor survival outcome in our analysis. Of the other factors related to BP activity that were not included in this meta-analysis due to insufficient data for statistical pooling, such as pruritus [
15,
29] and eosinophilia [
1,
15,
22,
29], no mortality difference was found by previous investigations. These findings suggest that inherent factors such as age and gender, as well as disease severity at the time of diagnosis, may not influence mortality as much as general condition and concomitant morbidities.
Our analysis corroborated previous finding that IIF is not a sensitive indicator of BP clinical activity [
18], but suggested that the presence of circulating antibodies regardless of detection method increased mortality risk, pointing to a predictive role of ELISA test. Unfortunately, since only three studies investigated survival prognosis based on ELISA result, we did not perform statistical pooling for this. To date, IIF plays a key role in the detection of circulating antibodies and is often used as the routine test. However, the circulating autoantibodies can also be detected by more accurate serological assays, including immunoblotting and ELISA [
18]. Of note, ELISA has been demonstrated as a useful tool to detect circulating antibodies in BP with its high sensitivity and specificity, particularly to recombinant BP180 antigen [
18]. Moreover, it has the ability to reflect the severity of bullous pemphigoid, presumably because autoantibody against BP180 is believed to be pathogenic in BP [
10]. It is noteworthy that this represents an intrinsic risk factor that correlates with BP mortality.
Severe concomitant morbidities at the time of diagnosis of bullous pemphigoid, such as neurological disorders, cardiac insufficiency, diabetes, and malignancies, increase the risk of mortality [
3,
12,
21]. Interestingly, there is a strong link between BP and neurological diseases. In this study, we evaluated the impact of two most common neurological diseases in BP patients, dementia, and stroke, on the survival of BP patients. Our pooled data indicated a roughly two-fold higher mortality risk for BP patients with dementia or stroke. The exact mechanism underlying the link between neurological diseases and BP has not yet been fully elucidated, but a putative autoimmune reaction against BP antigens in the brain, via a compromised blood–brain barrier, that cross react with the skin have been speculated as a possible cause [
20]. Neurological diseases may be responsible for the poor prognosis of some patients with BP because of their altered general condition. However, patients with neurodegenerative disorders are known to have increased blood–brain barrier permeability that may facilitate the generation and crossing of BP autoantibodies into the periphery, worsening skin damage, and clinical outcome [
8]. A recent study revealed that in a group of patients with Alzheimer’s disease but no bullous pemphigoid, their sera contained much higher level of BP180 autoantibody than neurologically healthy controls. Furthermore, increased levels of the autoantibody were associated with more severe dementia [
19]. Thus, it remains unclear whether neurological disease itself could increase the mortality risk, or the increased risk is conferred by functional impairment and decreased autonomy resulting from neurological disease. Future prospective studies should assess patient’s functional status at initial presentation and employ a multivariate analysis to answer this question.
Meta-analysis of some other prognostic factors was not conducted, because the data were too sparse or heterogeneous to combine analytically. Poor general condition, low karnofsky scale, or being bedridden has been repeatedly reported to be associated with unfavorable outcome [
3,
15,
32], with the exception of one study conducted by Lee et al. [
21], which did not find a significant association between karnofsky scale and BP mortality after multivariate analysis. Low serum albumin level and high erythrocyte sedimentation rate (ESR) were found to be associated with elevated mortality risk in BP [
22,
30]. High ESR may be a surrogate marker for more severe inflammatory status and more severe disease. Serum albumin level is related to patient’s nutritional condition, or underlying diseases, or extensive cutaneous erosions. However, these two factors are easily influenced by various factors, such as other acute and chronic illnesses, so they do not represent intrinsic biological variables of bullous pemphigoid.
Steroid is the first line therapy in both localized and generalized BP [
17]. High dosage of systemic corticosteroid or immunosuppressive agents correlates with higher mortality and increased side effects especially in patients at risk, as early deaths in BP patients are mainly of infectious or cardiovascular origin [
16]. However, we were not in an optimal situation to validly investigate this issue, because (1) treatment modalities had not been predetermined and controlled in most studies in our analysis; (2) the average dosage of steroid used in different studies varied greatly; (3) dosage and route of administration were largely determined by patients’ disease status; and (4) patients were often administered a combination of systemic steroid, topical steroid, and adjunctive immunomodulators.
Potential limitations need to be considered when interpreting the results of our meta-analysis. First, since most included studies were retrospective and conducted in referral centers, a potential unidentified confounding, information, and selection bias may exist, causing us to interpret our results with caution. Second, the follow-up time varied from 1 to 5 years among studies, which may influence the comparability of the results between component studies. Third, pooled results for some factors, such as mucosal involvement and positive IIF, were based on a limited number of studies. Therefore, we cannot rule out the possibility that insufficient statistical power is present. Finally, we could not evaluate all of the predictive factors for BP mortality in this meta-analysis.