Conflicts of interest The IVDK is sponsored by the cosmetic and chemical industry associations. W.U. has accepted honoraria for presentations or travel reimbursement from cosmetic industry associations for presentations given to these. A.S. is engaged occasionally as an ad hoc consultant for the cosmetic industry (associations), partly remunerated. J.S. and O.G. have no conflicts of interest.
‘Polysensitization’ (PS) is usually defined as contact sensitization to three or more unrelated haptens of the baseline patch test series. Despite PS being an important clinical phenotype indicating increased susceptibility to contact allergy, statistical approaches to analyse PS have hitherto been preliminary.
To apply an appropriate regression model for count data, namely, negative binomial hurdle regression, to a large set of clinical patch test data with the aim of estimating PS risk in more detail than previously achieved.
The detailed information provided by the hurdle model includes a separate estimation of an ‘increment factor’ quantifying the likelihood of further positive reactions, i.e. PS. Clinical data of 126 878 patients patch tested by departments comprising the IVDK (Information Network of Departments of Dermatology) network (www.ivdk.org) between 1995 and 2010 were included.
Regarding anatomical sites as exposure (surrogate), the axillae and the feet were found to be strong PS risk factors. Moreover, age was a strong PS risk factor, and less so, female sex. In comparison, atopic eczema and occupational dermatitis were less important risk factors. Single allergens contributed to PS to a varying extent.
The data presented point to some, very likely exposure-related, risk factors which need to be considered in future PS research, e.g. addressing the genetic basis for PS.
‘Polysensitization’ (PS) has become an important concept in the clinical epidemiology of contact sensitization. Generally, PS has been defined as the occurrence of three or more positive reactions to baseline series allergens. Regarding the definition and applications of the concept of PS, basically to identify persons who are apparently easier to sensitize than others, we refer to previous reviews on this topic.[1, 2]
To date, different statistical approaches have been used to examine the factors associated with PS, namely: (i) crude, bivariate analyses; (ii) logistic regression analysis; and (iii) nominal and ordinal regression analysis, using an (ordered) categorical outcome. The present analysis intended to utilize the maximum information available in the original count data (individual number of positive reactions to baseline series allergens). Therefore, a conceptually suitable multiple regression method was applied to a large collection of patch test data from the IVDK (Information Network of Departments of Dermatology). The aim was to identify risk factors for PS in more detail than previously possible, e.g. regarding anatomical sites associated with PS. The improved description of the PS clinical phenotype will, in turn, be useful in future in depth research trying to elucidate the genetic basis of (enhanced) susceptibility to allergic contact sensitization.
Material and methods
The IVDK (www.ivdk.org) is a clinical network dedicated to contact allergy surveillance in central Europe. Its methods have been described elsewhere. Briefly, patch test data are collected along with a standardized patient history and transferred to the data centre in Göttingen, Germany, twice a year in an anonymous format. Only those data complying with internal quality standards are included in scientific analyses such as the present one. All participants of the IVDK are members of the German Contact Dermatitis Research Group (DKG) and comply with international patch test guidelines, which have been refined nationally. The German baseline series maintained by the DKG currently comprises 30 allergens. As PS implies sensitization to unrelated allergens, positive reactions to related allergen preparations (in terms of the same, or a closely related substance being present in both preparations) were counted as one positive reaction. This applies to the following pairs of allergens: thiuram mix and zinc diethyldithiocarbamate, mercaptobenzothiazole and mercapto mix, fragrance mix II and hydroxyisohexyl 3-cyclohexene carboxaldehyde (HICC, e.g. Lyral™), methyldibromoglutaronitrile and methyldibromoglutaronitrile + phenoxyethanol (Euxyl™ K 400).
The number of allergens comprising the baseline series changed slightly during the study period (Table 1). Four allergens in particular (paraben mix at 15%, historically, and sandalwood oil, ylang-ylang oil and Jasmine absolute, recently) had been tested in only about 1500 patients. While the (few) positive results were included in the calculation of PS, detailed results are not presented in Table 1 as these would be very imprecise. Generally, patch test readings on day 3 or, if these were not available, from day 4 were analysed. Positive reactions of all grades, + to +++, were aggregated to an outcome ‘positive’ as opposed to ‘nonpositive’ reactions, which comprise all other reactions (negative, doubtful, irritant). For the present analysis, patch test data from 65 centres in Germany, Austria and Switzerland, collected from 1995 to 2010 were used, amounting to a data pool of 142 470 patients. After elimination of observations with incomplete data, mostly due to missing data on the anatomical site of dermatitis, 126 943 patients remained in the final clinical sample.
Table 1. Prevalence of positive reactions to single baseline allergens and additional positive reactions (grouped) in patients patch tested in the IVDK network 1995–2010
‘0 add’, proportion of patients without additional reactions to baseline series allergens among those with a positive reaction to the respective allergen; ‘1 add’, proportion of patients with one additional reaction to a baseline series allergen among those with a positive reaction to the respective allergen; ‘2+ add’, as for 1 add, with two or more additional positive reactions. Ordering of allergens is by decreasing proportion of 2 or more additional reactions. Sodium lauryl sulfate, an ‘irritant control’ and petrolatum (vehicle control) not considered. aMixture of 5-chloro-2-methyl-isothiazol-3(2H)-one and 2-methylisothiazol-3(2H)-one.
For statistical analysis, the statistical package R (version 2.14.1, http://cran.r-project.org/) was used. The requirement for a regression model was that it: (i) uses the full information in the data, i.e. the original count data; and (ii) offers a separate estimation of the risk of additional positive reactions and the risk of being sensitized at all. The negative binomial hurdle regression model met these requirements and was therefore used (R package ‘pscl’). This method separately estimates the risk of being sensitized at all (which may depend on the selection process until patch testing) and the risk of having several contact allergies, i.e. to be polysensitized. Risk factors included a priori in the model, based on previous, related analyses, were age, sex, atopic eczema (past or current), occupational causation of dermatitis, duration of application of the patch test (48 vs. 24 h) and anatomical site of dermatitis. Additionally, the number of individually tested allergens was controlled for, as this evidently is associated with the outcome (firstly, the length of different versions of baseline series differs to some extent; secondly, allergens are occasionally omitted on a case by case basis).
Table 1 lists the single allergens comprising the German baseline series, the number of patients tested and the crude proportion of positive reactions on the left hand side. While the vastly varying prevalence of positive reactions is well known and commented on elsewhere, the contribution of the different allergens to PS shown in the three right hand side columns is a novel finding, relevant for an understanding of PS. For example, nickel and thiomersal contribute relatively little, with > 40% isolated reactions due to a uniform exposure unrelated to other allergens (piercing and vaccines, respectively). In contrast, cobalt is often coupled (with nickel). A number of other, mostly fragrance or medicament, allergens show a low share of around 20–25% of isolated reactions and thus a larger contribution to PS.
The distribution of the outcome variable is shown in Figure 1, together with the counts predicted by the negative binomial hurdle regression model. As a first descriptive approach to examining the relationship between PS and the set of explanatory factors considered, the relative percentages of the number of positive reactions, aggregated into three categories for better readability and comparability with previous analyses, in the strata of explanatory factors are shown in Table 2. Already this simple bivariate analysis reveals certain patterns, such as an increasing proportion of PS (‘3+ ’ reactions) with age. Specific sites may enhance the risk of sensitization (axillae and feet due to occlusion, leg dermatitis with pre-existing inflammatory milieu), which adds to the general risk of being sensitized, thus increasing the probability of PS.
Table 2. Proportions of (poly) sensitized patients, categorized as 0 (no sensitization at all), 1/2 (sensitization to one or two allergens) and 3+ (sensitization to three or more allergens, corresponding to the conventional definition of polysensitization) in subgroups of patients, i.e. strata of potential risk factors
Data from the IVDK network, 1995–2010.
< 34 years
≥ 62 years
Previous or current atopic eczema
2-day duration of patch test application
Table 3 shows the results of the mutually adjusted estimates from the negative binomial hurdle model. The exponentiated coefficients shown can be interpreted as odds ratio (OR; ‘threshold risk’) in the zero part and as relative risk (RR; ‘increment factor’) in the count part. Thereby, the former represent the risk (as OR) of being sensitized at all, the latter the risk for one more additional positive reaction. Taking, for example, sex as an explanatory factor, a female patient has an expected number of positive reactions that is 12% higher than in a male patient with the same characteristics regarding age, anatomical site, etc. For instance, if a man has two positive reactions, the corresponding number is 2·24 for a woman. Evidently, these noninteger numbers of positive reactions cannot occur in reality; they are just theoretical estimates of the risk associated with the respective factor. Considering the ‘threshold risk’, female sex increases the odds by 71%. Compared with patients aged < 34 years, patients aged between 34 and 47 years show the greatest odds of being sensitized at all, namely, a 26% increase. This ‘threshold risk’ decreases again in the two oldest age groups, whereas the ‘increment factor’ steadily increases. This divergent characteristic is discussed later.
Table 3. Estimates of odds ratio (OR) and relative risk (RR) obtained from the hurdle model
Data from the IVDK network, 1995–2010. CI, confidence interval; ‘Zero part’, the OR denotes the odds of being sensitized at all (to at least one baseline series allergen) as a ‘threshold risk’; ‘Count part’, the RR for one more additional positive reaction(s), respectively, associated with the risk factor as an ‘increment factor’. aNot listed, but included in the regression analysis ‘Other’, i.e. all remaining sites; meaningful interpretation of the OR estimate not possible.
A 2-day duration of the patch test, compared with 1 day, is associated neither with a relevant increase of the OR (8%) nor with a significant ‘increment factor’. Regarding the anatomical site of dermatitis, it should be mentioned that the actual estimates, and the departure from unity of the corresponding confidence interval, depend on the choice of the reference group (in our analysis: trunk), which is arbitrary in this case of a nominally scaled variable. However, the ranking of risk estimates is interpretable and shows that leg, foot and axilla are associated with a high number of positive patch test reactions.
Often, simple statistical models are used to evaluate medical data, as they can be applied by a broad range of researchers, and can often be easily interpreted. On the other hand, application of too simple statistics: (i) may lead to erroneous conclusions, e.g. by not accounting for confounding; and (ii) may not exploit fully the information available in the data, e.g. by unnecessarily aggregating outcome or explanatory variables. In this unprecedented analysis, we have applied a multifactorial regression model, which is ideally adapted to the particular distribution of our outcome variable, accounts for several explanatory factors or confounders, exploits the full information available in the data, and offers a separate estimation of an ‘increment factor’ representing PS, respectively, to examine clinical factors associated with PS.
The current definition of PS is ‘sensitization to three or more unrelated allergens included in the baseline series’. It should be noted that this is, to some extent, an arbitrary definition, which has become a convention. The overall yield of the baseline series (and, accordingly, the probability of diagnosing PS) will vary with the appropriateness of continually adapting this series to the currently most important contact allergens. Evidently, additional positive reactions can occur when testing additional test series, adapted to the individual patient's history. However, the scope of patch testing, and, with it, the probability of diagnosing (additional) contact sensitization, would get too diverse if such individualized testing is considered. Hence, only baseline series results are included in the PS concept. Moreover, no formal evaluation of an ‘optimum cut point’ has been undertaken so far, and would probably be a futile undertaking anyway, in view of the lack of an external gold standard such as a valid genetic marker. Furthermore, the choice of a reference group, namely, patients without positive reactions, or patients with just one specific contact sensitization to a baseline series allergen, is not self-evident and has been handled differently in the few different studies addressing PS as the outcome. Hence, there seemed no convincing reason to adhere to a certain categorical definition.
Initial results show differences between more ‘solitary’ allergens such as nickel, and more ‘co-reactive’ allergens such as cetearyl alcohol, parabens and oil of turpentine, in line with a previous analysis. Furthermore, those allergens with a low proportion of no additional reactions can be considered as (very) weak allergens, which have already been shown to be associated with PS. Hence, PS is a phenomenon structured by certain groups or clusters of haptens.
When interpreting the results presented in Table 3, it should be kept in mind that the novel approach of deriving two separate estimates is difficult to compare with previous results in quantitative terms. Previous analyses comparing levels of a categorical PS outcome with zero positive reactions as reference combine, in a way, both of our components in one model. Other analyses comparing an arbitrary level (e.g. one reaction) to another level (e.g. three or more reactions), only address in the limited, categorical framework, the second component of our model. Notwithstanding this interpretational difficulty related to different analytical concepts, it is interesting to compare our results with previous findings.
Sex seems to be an important risk factor for sensitization in general but not for an exceptionally high number of positive reactions. Women apparently are found to be sensitized more often than men, as previous studies have already shown.[15, 16] It is, however, not entirely clear whether this observation is explained by constitutional differences between the sexes, or by women being more intensely exposed, e.g. occupationally or to consumer products, or by both factors. Regarding age, the chance of being sensitized and the risk of having multiple contact allergies is smallest for people aged < 34 years, our reference group. An explanation for this observation could be the shorter exposure to both consumer and working-life allergens, notwithstanding some allergen exposures related particularly to young age, such as, formerly, nickel. Interestingly, the ‘threshold’ estimate decreases again for the oldest subgroup to almost the value of the youngest (reference) group. In contrast, the steady increase of the risk of multiple contact allergies with increasing age most likely reflects the effect of the cumulative exposure to a multitude of allergens during the lifespan, despite an apparently decreasing capacity to be newly induced (sensitized).
Previous or current atopic eczema is moderately associated with PS, corresponding with findings from previous, more general analyses.[19, 20] Occupational dermatitis is similarly associated with PS, but more strongly than atopic eczema with having at least one contact allergy. A plausible explanation for this phenomenon is the often intense exposure to allergens during working time, sometimes under occlusive conditions and with pre-existing skin irritation, which further facilitates sensitization.
Patch tests are applied either for 1 or 2 days. In previous studies, more irritant and allergic reactions were seen after 48 h of allergen exposure, also in relation to the proportion of positive (allergic) reactions. According to a review a 2-day application is preferable. Reassuringly, according to our multiply adjusted results, the impact of exposure time is minute regarding the ‘threshold risk’, and negligible (nonsignificant) concerning PS.
Previous studies have already shown an association between anatomical sites and sensitization to a number of single allergens (e.g. see references  and ). Regarding PS, using the above-mentioned categorical definition, Carlsen et al. identified associated sites, albeit employing coarser granularity. In the ranking of risk associated with anatomical site in the categories we used, the axilla (the smallest subgroup of patients) is most strongly associated with a ‘threshold risk’. This is not surprising as the axilla is a locus minoris resistentiae due to semiocclusive conditions and repeated microtrauma by shaving. Thus the axillae are predisposed to sensitization, particularly to fragrance mixtures used in deodorants,[26, 27] which may, as mixtures, additionally promote sensitization and elicitation. The presence of several fragrance allergens in the baseline series (the two mixes, Myroxylon pereirae, HICC and oil of turpentine) may be mainly responsible for a moderately increased PS risk. Pre-existing leg ulcer or stasis dermatitis strongly predisposes to the acquisition of any contact sensitization. The longer a chronic leg ulcer is present, the more positive reactions are seen in the patient. Our results show strong associations of leg dermatitis between both the ‘threshold risk’ and PS, in line with previous findings. It must be stressed that the increase in risk we observed is not attributable to age, which is a strong confounding factor, as our analysis had been adjusted for age. The hand is, in our model, the only site showing a greater impact on PS than on ‘threshold risk’ – acknowledging that the two different risk estimates are not directly comparable numerically. The hand is probably exposed to the broadest array of different potential allergens. In some cases, it is the only exposed site (as often in an occupational setting – included as explanatory factor); in other cases, it is coexposed along with other sites (as, for instance, when applying body lotions, topical medications or hair cosmetics). Hence, the strong association with a vast number of allergens, ultimately leading to being a strong risk factor for PS, can be well explained. In line with our results, Carlsen et al. found an association between hand dermatitis and PS. The feet have not been considered separately in previous analyses of PS. However, in view of the importance of this site presently observed, both in terms of ‘threshold risk’ and PS, its consideration seems worthwhile. Multiple allergens (e.g. in shoes and socks, such as dyes, glues, chromium as a tanning agent, and others) may play some role as sensitizers, in combination with aggravating factors such as occlusion, sweating and friction. Although the face is an important site, with a partly characteristic spectrum of allergens, facial contact dermatitis is not related to PS, as in the study of Carlsen et al. Other anatomical sites considered in our study, but not in previous analyses, are not associated with a markedly increased or decreased risk of PS.
In conclusion, our results provide an accurate description of exposure factors related to the PS clinical phenotype. This information should be taken into account in further research addressing the genetics underlying PS. Moreover, longitudinal analyses should address the development of patients’ reactivity from single or ‘oligo’-sensitization to PS, which is presumably a staggered phenomenon.
The following centres contributed their data to the analyses (in alphabetical order): Aachen (H. Dickel, S. Erdmann, C. Schröder, J.M. Baron), Aarau (J. Grabbe), Augsburg (A. Ludwig), Basel (A. Bircher), Berlin B.-Frank. (B. Tebbe, R. Treudler, M. Worm), Berlin BWK (A. Köhler), Berlin Charité (B. Laubstein, T. Zuberbier, M. Worm), Berlin UKRV (J. Grabbe, T. Zuberbier), Bern (D. Simon), Bielefeld (I. Effendy), Bochum (Ch. Szliska, M. Straube, H. Dickel), Bochum BGFA (M. Fartasch), Buxtehude (P. Hausenblas, E. Wrage-Brors, P. Große-Hüttmann), Dessau (U. Lippert, A. Jung), Dortmund (P.J. Frosch, B. Pilz, C. Pirker, R. Herbst, K. Kügler), Dresden (G. Richter, R. Aschoff, P. Spornraft-Ragaller, A. Bauer), Dresden Friedrichstadt (A. Koch), Duisburg (J. Schaller), Erlangen (K.-P. Peters, T.L. Diepgen, M. Fartasch, V. Mahler, M. Hertl), Essen (H.-M. Ockenfels, U. Hillen), Falkenstein (H. Schwantes), Freudenberg (Ch. Szliska), Gera (J. Meyer, H. Grunwald-Delitz), Göttingen (J. Geier, Th. Fuchs), Graz (B. Kränke, W. Aberer), Greifswald (M. Jünger), Halle (G. Gaber, D. Lübbe, B. Kreft), Hamburg (D. Vieluf, M. Kiehn, R. Weßbecher, E. Coors), Hamburg Dermatologikum (K. Reich, V. Martin), Hamburg BUK (C. Schröder-Kraft, K. Breuer, U. Seemann), Hannover (T. Schaefer, Th. Werfel), Heidelberg (A. Schulze-Dirks, M. Hartmann, U. Jappe, K. Schäkel), Heidelberg AKS (H. Dickel, E. Weisshaar, T.L. Diepgen), Heilbronn (H. Löffler), Homburg/Saar (P. Koch, C. Pföhler), Jena (M. Gebhardt, A. Bauer, W. Wigger-Alberti, M. Kaatz, S. Schliemann), Kiel (J. Brasch), Krefeld (A. Wallerand, M. Lilie, S. Wassilew), Leipzig (R. Treudler), Lippe Detmold (St. Nestoris), Lübeck (J. Kreusch, J. Grabbe, I. Shimanovich, U. Jappe), Magdeburg (U. Beier, B. Stötzel, R. Vetter), Mainz (D. Becker), Mannheim (Ch. Bayerl, D. Booken, H. Kurzen, C.-D. Klemke), Marburg (I. Effendy, H. Löffler, M. Hertl, W. Pfützner), Minden (R. Stadler), München LMU (B. Przybilla, P. Thomas, T. Oppel, T. Schuh, R. Eben), München Schwabing (M. Agathos, K. Ramrath, M. Georgi, G. Isbary, K. Ramrath), München TU (J. Rakoski, U. Darsow), Münster (B. Hellweg, R. Brehler), Nürnberg (I. Müller, D. Debus, A. Bachtler), Oldenburg (M. Padeken), Osnabrück [W. Uter, S.M. John, H.J. Schwanitz (deceased), N. Schürer, H. Dickel, Ch. Skudlik], Rostock [H. Heise (deceased), J. Trcka, Ch. Schmitz, M.A. Ebisch], Stuttgart (G. Nist, T. Plaza), Tübingen (G. Lischka, M. Röcken, T. Biedermann, J. Fischer), Ulm [H. Gall (deceased), G. Staib, P. Gottlöber], Ulm BWK (H. Pillekamp), Ulm Uni (G. Staib, J. Weiss, R. Hinrichs), Wuppertal (O. Mainusch, J. Raguz), Würzburg (A. Trautmann, J. Arnold), Zürich (B. Ballmer-Weber), Zwickau (B. Knopf, D. Teubner).