Prevalence of inherited ichthyosis in France: a study using capture-recapture method

Background Inherited ichthyoses represent a group of rare skin disorders characterized by scaling, hyperkeratosis and inconstant erythema, involving most of the tegument. Epidemiology remains poorly described. This study aims to evaluate the prevalence of inherited ichthyosis (excluding very mild forms) and its different clinical forms in France. Methods Capture – recapture method was used for this study. According to statistical requirements, 3 different lists (reference/competence centres, French association of patients with ichthyosis and internet network) were used to record such patients. The study was conducted in 5 areas during a closed period. Results The prevalence was estimated at 13.3 per million people (/M) (CI95%, [10.9 – 17.6]). With regard to autosomal recessive congenital ichthyosis, the prevalence was estimated at 7/M (CI 95% [5.7 – 9.2]), with a prevalence of lamellar ichthyosis and congenital ichthyosiform erythroderma of 4.5/M (CI 95% [3.7 – 5.9]) and 1.9/M (CI 95% [1.6 – 2.6]), respectively. Prevalence of keratinopathic forms was estimated at 1.1/M (CI 95% [0.9 – 1.5]). Prevalence of syndromic forms (all clinical forms together) was estimated at 1.9/M (CI 95% [1.6 – 2.6]). Conclusions Our results constitute a crucial basis to properly size the necessary health measures that are required to improve patient care and design further clinical studies.


Background
Inherited ichthyoses represent a large group of monogenic Mendelian disorders of cornification due to mutations in genes involved in skin barrier function [1]. Skin changes, affecting all or most of the tegument, are characterized by scaling, hyperkeratosis or both. Erythema is inconstantly present. According to a recent ichthyosis expert consensus, the classification should be clinically based and distinguish between syndromic and nonsyndromic ichthyoses [2].
Except for very mild forms of ichthyosis, the disease was demonstrated to have a significant impact on quality of life (QOL) [3][4][5][6][7]. During the past few years, much progress has been achieved in describing the molecular basis of these disorders and in establishing genotype-phenotype correlations [8]. However, prevalence remains poorly described [9][10][11][12][13][14][15][16] and no patient registration exists. Understanding epidemiology of rare skin disorders such as ichthyosis is one of the main objectives advocated in the Plan Maladies Rares defined by the French Health ministry [17].
The objective of our study was to estimate the prevalence of inherited ichthyoses having a significant impact on QOL (excluding accordingly very mild forms) and to determine the prevalence of their different clinical forms. Patients suffering from such inherited ichthyoses are primarily diagnosed by private or hospital physicians and subsequently usually followed in hospital. Nevertheless, a "hospital list" may not be exhaustive to enumerate such patients, since some of them, who could have been demotivated by the lack of effective treatment for example, might have interrupted their medical care and thus are "out of the hospital healthcare system". It is therefore necessary to identify these patients via additional lists of recruitment, hence the choice of capture-recapture method [18][19][20][21][22][23][24].
Capture-recapture models are commonly used for estimating prevalence in epidemiological studies when information comes from incomplete lists [18,19]. Estimation is based on the number of cases recorded on each list, as well as the overlapping of cases between the lists. This thereby allows for estimation of the number of cases of the target population appearing on no list. The standard statistical approach uses log-linear models. They allow for dealing with possible dependencies between lists, but also include, in particular, the independent model which assumes independence between lists [20,21]. The implementation of such an approach requires at least 3 sources of data.

Methods
The study protocol was reviewed and approved both by an independent health ethics committee in Toulouse, France and by the CNIL, the French National authority for registers.
The target population consisted of patients suffering from inherited ichthyosis with a confirmed diagnosis performed by a physician at a given time of their illness. We excluded very mild forms that can be defined as ichthyoses with a very mild clinical severity (visual analogue scales (VAS) for erythema and scaling < 1.5/10, in the absence of systemic therapy by acitretin). In practice, these very mild forms correspond to ichthyosis vulgaris or ichthyoses with similar severity.
According to statistical requirements, 3 lists of patients were selected: patients registered in reference/competence centres that are experts for rare skin diseases (L1), members of the French association of patients with ichthyosis (named AIF) (L2) and members of the Facebook social network: "L'ichtyose qu'est-ce que c'est?" (L3). The study was conducted in 5 French areas (Aquitaine, Bourgogne, Midi-Pyrénées, Pays de la Loire, Provence Alpes Côte d' Azur (PACA)) during a closed period of 6 weeks (from June 1 st 2011 to July 15 th 2011). Patients were contacted either by phone (L1, L2) or by email (L3) to obtain their consent. Overlaps between lists were identified by checking the first 3 letters of the surname, the first 2 letters of the first name and the date of birth. In order to confirm the population prerequisite closed in time and space, all patients were also asked about their zip code and whether they resided in the selected area for a minimum of 6 weeks (without migratory movements).
For the patients recruited via L1, the diagnosis of the clinical form of ichthyosis and the severity were determined by the clinician in charge. The patients recorded via L2 or L3 (without overlapping with L1) were asked to fill-in a questionnaire exploring ichthyosis characteristics (including VAS for erythema and scaling), which was then reviewed by a board of ichthyosis experts to define the clinical form and the severity (data not shown).
The number of patients suffering from ichthyosis was estimated by considering a set of log-linear models, taking into account possible interactions between lists. In addition to area, individual covariates -such as gender, age (older than 15 years or younger) and grade of severity (mild/moderate/severe/very severe) -were introduced in the model in order to improve the prevalence estimation [25]. A selection model procedure was implemented using the GENMOD procedure of SAS: the best model being the one with the maximum weighted Akaike Information Criterion (wAIC) [26]. The selected model allowed for estimation of the number of ichthyosis patients at several levels: estimation for each area and for the pooled areas, as well as estimation by age group and by severity. Estimated prevalence was finally deduced from the estimated number of ichthyosis patients and the French 2008 census: for each area (regional prevalence) and for the pooled areas (global prevalence), which can be considered as national prevalence, subject to representativeness of the selected areas.
With regard to clinical form, a separate ad hoc statistical procedure was implemented (clinical form cannot be introduced as covariate in the log-linear model because of a very low number of detected patients for some clinical forms). The prevalence of each clinical form was estimated by applying the distribution of clinical forms observed among the reported cases to the estimated global prevalence. This ad hoc procedure is based on the assumption that the capture probability depends on the clinical form mainly through the severity grade, which seems reasonable from a dermatological point of view.

Three-source observed data
A total of 119 patients suffering from ichthyosis were identified. The observed frequencies according to presence or absence on each list for each area are presented in Table 1. Data was unbalanced: the majority of patients were identified via L1, whereas very few patients were recorded via L3. Overlaps were rather rare and mostly observed between L1 and L2. No patient was captured simultaneously in the 3 lists. Furthermore, some geographic disparities were observed concerning the recruitment of the patients.

Estimation of the number of patients Global estimation (pooled-areas level)
Among all log-linear models considered (Table 2), the independent sources model with 3 covariates effects was selected: it assumes that the 3 sources are independent, and that the capture probabilities depend on each source and on the 3 individual covariates (area, age and severity grade). Estimation was based on data collected from 4 areas (Aquitaine, Midi-Pyrénées, PACA and Pays de Loire), Bourgogne area being excluded because of non-valid estimations due to very low counts. This model gave a global estimation of 191 patients (95% CI [157-253]) for the 4 selected areas.

Estimation according to individual covariates
From this model, we also estimated the number of ichthyosis patients by age group (  [10][11][12][13][14][15][16][17]). The probability of appearing in at least one list increased with disease severity: it was estimated that 95% of very severe patients (grade 4) were detected in the study vs. only half of mild patients (grade1).

Estimation for each area
The estimated number of ichthyosis patients for each area is presented in Table 4. Bourgogne data were excluded from analysis but can be included in models without covariate. When comparing estimates from these models on 4 or 5 areas, we deduced that the estimated number of patients in Bourgogne was around 9 (no confidence interval available with this ad hoc approach).

Estimated prevalence of the different clinical forms of ichthyosis
The diagnosis of the clinical form of ichthyosis was performed by the clinician in charge for 85.7% (102/119) of Table 1 Numbers of patients reported in the three lists, in each of the 5 selected areas

Discussion
This is the first study investigating prevalence of inherited ichthyosis (excluding very mild forms) in France. This prevalence was estimated at 13.3/M (95% CI, [10.9-17.6]), with more than half of the patients suffering from ARCI.
Capture-recapture method was appropriate for estimating prevalence of rare diseases. Indeed, cross-sectional studies are the main alternative to estimate prevalence but may fail in case of rare disease because, even in large samples, it is possible to detect no case [27]. Requirements of the capture-recapture method were respected [18]. The target population was closed (no entry or loss of patients during the study period); individuals have been correctly recorded; there was no matching error; any individual of the target population had a strictly positive probability to appear on any list.
The choice of our 3 lists was based on the followings points: patients are often seen even once in expert centres for rare skin disease (L1). The association AIF (L2) is the only patient support group in France and registers a significant number of patients (approximately 250 members). The choice of a social network (L3) was based on patients' enthusiasm for searching information or communicating via internet (more than 100 members). Recruitment via L2 and L3 (17 non-hospital patients: 14.3%) was not negligible which confirmed the appropriateness of the capture-recapture method and the choice of the lists. The choice of the 5 areas was based on the presence of a reference/competence centre. These areas can be considered as representative of France since they are scattered throughout the country and cover more than 20% of the population.
From a statistical point of view, sparse and unbalanced data induced some difficulties. Indeed it was not possible to include the clinical form in our selection model procedure because the statistical procedure implemented by SAS did not provide valid estimates due to computational difficulties. These statistical difficulties were solved by implementing ad hoc approaches. Other statistical approaches which require some specific theoretical developments are possible and could be proposed in future studies. To complete our statistical analysis, the pseudo-R 2 proposed by Mc Fadden [28] was calculated: 74% of data variability was explained by the selected model, highlighting its good fitness and its explanatory power.
One limitation of the study could be the use of a selfassessment severity scoring by patients from L2 and L3 who could not be seen by the clinicians involved in the study. Nevertheless, all diagnoses were unequivocally determined by the expert board. In addition, our study did not provide an estimation of the prevalence of the entire ichthyosis population. This could be achieved in the future by the creation of well-designed register involving also private physicians.
In the literature, only a few studies have specifically focused on ichthyosis's prevalence. Earliest studies were conducted outside Europe (Saudi Arabia and Tunisia) and concluded to prevalence which seems different than European one because of a high level of consanguinity [9][10][11][12]. In Europe, two recent Spanish studies have dealt with ichthyosis prevalence [13,14]. Contrary to our study, only ARCI were concerned. The first one found a prevalence of ARCI consistent with ours, although slightly higher (8.2/M) but the method used for estimation was not clearly described [13]. The second one was an epidemiological study (144 patients) that used the same capture-recapture method [14]. ARCI's prevalence was almost identical to ours, with a similar proportion of LI/CIE/minor variants: 7.2/M (CI 95% [5.7 -9.7]; 62.5/30.6/7%) for the Spanish estimation vs. 7/M (CI 95% [5.7 -9.2]; 64/28/8%) for our study. In addition, estimates of ARCI's prevalence in the Spanish population depended on the age group (the younger the age group, the higher the prevalence) with an ARCI's prevalence of 16.2/M (CI 95% [13.3 -23.0]) for the patients younger than 10 years. This age effect was also found in our estimates (23/M (95% CI [19][20][21][22][23][24][25][26][27][28][29][30][31]) for patients younger than 15 years old (12.1/M (95% CI [10.2 -16.5]) when only considering ARCI). In the United States, incidence of ichthyosis during the first year of life (excluding common forms which are not diagnosed at birth) was estimated to 6.7 per 100.000 from large administrative claims databases [15].
With regards to keratinopathic ichthyosis, a recent Japan survey had estimated the number of patients to 55 (CI 95% [35 -75]) which corresponds to a lower prevalence (0.44/M (CI 95% [0.28 -0.59])) than ours. Nevertheless, Japanese patients seen only in Dermatology departments were concerned [16]. With regard to XLI, some studies had concluded to a substantial prevalence, possibly excessive, that ranges from 1 case per 10 000 to 1 case per 2 000 males (large ratio of 5:1), depending on the study population and the method of assessment [29][30][31][32]. Our XLI prevalence was lower. It is therefore difficult to conclude formally about the real prevalence of XLI. Our prevalence may be underestimated and could mostly concern patients who are suffering from moderate to very severe forms. Mild forms of XLI are probably less bothersome and these patients may not consult a hospital or be members of patient's association or social network. This is in accordance with the small number of patients (n = 73) diagnosed as XLI during the 10 past years in France using enzymatic blood test (personal data provided by the only French diagnosis centre).

Conclusion
This is the first French epidemiological study on ichthyosis. It focused on ichthyoses with significant impact on QOL, usually followed in hospital and for which medical, social or research progresses are needed. The obtained results constitute an essential basis for properly sizing the necessary health measures such as access to molecular diagnosis or design of clinical studies, especially in the domain of therapeutics in which there is a true lack.