Hand eczema is caused by many factors. The point prevalence of hand eczema is approximately 4% in adults and the annual prevalence is around 10%.1–3 The frequency of contact allergy diagnosed on the basis of positive skin tests ranges from 23% to 62% in patients with hand eczema, while the rate of positive test results of current clinical relevance ranges from 20% to 25%.4,5 Guidelines based on expert opinion all recommend patch testing but do not specify when or under which circumstances they are indicated. The only established criterion is that the condition be chronic, defined by the presence of symptoms lasting for varying periods of time (4, 6, or 12 weeks) during which hand eczema remains poorly controlled even with appropriate treatment.1–6

Predictive models of prognosis or risk rely on a combination of factors with different degrees of association with an event, making it possible to calculate the likelihood that an individual will experience it.7 These models are useful for making clinical decisions about treatment and have been widely used in dermatology. An example is a model for scoring pigmented lesions during diagnosis.8

Because a positive patch test reaction to an allergen of current relevance indicates a need to warn a patient with hand eczema to avoid the substance implicated, these tests are extremely important for managing therapy. We have found no predictive models for hand eczema, or for contact eczema in general, that name a positive patch test reaction of current relevance as the outcome of interest.

We aimed to develop a model to predict a positive patch test reaction to a substance of current clinical relevance for patients with eczema located only on the hands.

Table 1 summarizes the main patient characteristics. We observed 467 patch test reactions in 262 patients. At least 1 positive result was observed in 66.03% of the patients. The allergens that provoked the largest numbers of reactions were nickel (89), combined methychlorolisothiazolinone/isothiazolinone (27), methylisothiazolinone (18), ethylene glycol dimethacrylate (18), and hydroxyethyl methacrylate (17). Positive reactions of current clinical relevance were found in 28.6%. The most common ones are summarized in Table 2, which shows that acrylates were clearly the most frequent culprits. Patient occupations are shown in Table 3.

Predictive Model

The variables confirmed to be predictors on multivariate analysis were employment as a hairdresser–aesthetician, male sex, and patient history of atopic dermatitis. These variables were entered as present (1) or absent (0). The model deduced by logistic regression (Table 4) was as follows: 2.316×(hairdresser–aesthetician)−1.792×(atopic dermatitis)−0.601×(male sex). A patient's score was positive (presence of a currently relevant reaction) if it was higher than the cut-point K.

Table 4.

Factors in the Linear Logistic Regression Model.a

	Coefficient (SE)	AICb	Odds Ratio (95% CI)
Independent term	−0.708 (0.248)	–	–
Atopic dermatitis	−1.792 (0.846)	199.1	0.167 (0.032; 0.874)
Male sex	−0.601 (0.374)	195.4	0.548 (0.264; 1.140)
Hairdresser–aesthetician	2.316 (0.568)	213.7	10.1 (3.333; 30.8)

a

Model developed with the training data set (n=175).

b

The AIC value shown reflects the elimination of the variable; the AIC for the full model is 194.8. Note that the model's fit becomes poorer if any variable is removed. The algorithm selected the predictors using the AIC and the full enumeration.

Abbreviation: AIC, Akaike information criterion.

The model displayed moderate discrimination, with an AUC-ROC of 0.78 (95% CI, 0.66–0.90 (Fig. 1). The optimum cut-point according to the criterion of proximity to the upper right corner was K=−0.3. The properties supporting the model's ability to identify cases (discrimination) are shown in Table 5.

Figure 1.

Area under the receiver operating characteristic curve (AUC-ROC) for the model applied to the validation data set. The numbers shown on the curve correspond to cut-points. The optimum cut-point that satisfies the criterion of proximity at the top right corer was K=−0.3.

(0.11MB).

Table 5.

Properties of the Model at the Cut-point K=−0.3.a

Parameter	Estimate	95% CI
Sensitivity, %	73.7	48.8–90.9
Specificity, %	60.3	47.7–72.0
Positive predictive value, %	34.1	25.8–43.6
Negative predictive value, %	89.1	79.0–94.7

a

Values were obtained for a prevalence of currently relevant reactions of 28.6%, using the validation dataset (87 patients).

Predictive models are highly useful for clinical decision-making. Patch tests are done to detect sensitization to a product that is causing contact dermatitis. It would therefore be useful to have a predictive model that identifies patients who are most likely to have a positive patch test reaction of current clinical relevance. Our model relies on the presence or absence of 3 predictors: employment as a hairdresser–aesthetician, male sex, and a history of atopic dermatitis. These 3 factors were assigned optimal weights by logistic regression. The proposed predictive model, validated in a second data set different from the one used to develop the model, had an AUC of 0.78. For the selected cut-point (K=−0.3), the negative predictive value was 89.1%. Thus, when a patient has a score under −0.3, there is only a 10.9% chance of positive patch test findings of current relevance.

For example, if a non-atopic woman who works as a hairdresser–aesthetician is the patient, her score would be calculated as follows: 2.316×1−1.792×0−0.601×0=2.316. Since the resulting score is greater than −0.3, patch testing would be indicated with no need to wait to see her response to standard treatment with topical corticosteroids and emollients. The score for a man with atopic dermatitis who did not work as a hairdresser, on the other hand, would be calculated as follows: 2.316×0−1.792×1−0.601×1=−2.393. Given this score, a more conservative approach of prescribing hand care and waiting to see the response before testing would be advisable. There are certain constraints on applying predictive models, and the criteria for including patients when developing one must be made clear. We excluded patients who had eczema in other parts of the body in addition to the hands in order to attenuate an effect of clinical variability. A predictive model is valid for use in the study population or one of similar characteristics, so our model is not applicable to other dissimilar populations. The small size of our patient sample, the limited number of clinical variables analyzed, and the preponderance of patients who worked as hairdresser–aestheticians and who were sensitized to acrylates are limitations of our study. The lack of other clinical data stored in our database prevented us from including other variables, such as clinical type,5,13,14 which would have to be taken into account in more exhaustive multicenter studies.

The only model developed to predict hand eczema, to our knowledge, was published by Meding and Swanbeck15 in 1990. Their model was based on a set of clinical variables. We have found no study attempting to predict currently relevant positive patch test results, nor any other predictive model specifically for hand eczema. Most studies on hand eczema describe clinical variables that are associated with greater or lesser risk for allergic contact dermatitis and positive patch test findings. However, most of them do not take into consideration whether or not the positive findings are of current relevance in the patient's life.4,5,13,16–19

The rate of positive reactions was 66.03% in our data set, and the 28.6% that were of current relevance were mainly in women, consistent with previous studies.4,5,13,14,16,20

The association between hand eczema and occupational risk factors is well established and can be seen in over half of patient cohorts in Europe.5 However, only a limited number of studies link the frequency of positive patch test reactions to specific occupations.17–19 In addition, the occupation that predominates in studies differs according to population. Employment as a hairdresser–aesthetician was the main risk factor for contact dermatitis evidenced by a currently relevant positive test result in our study. This association was largely the result of the nature of the study population itself. The association between hand eczema and occupational exposure to acrylates in hairdressers has been described with ever increasing frequency in the last decade.21–23

The prevalence of atopic dermatitis in our study was 12.6%, and the presence of this factor was inversely associated with having a positive reaction of current relevance. Although studies are not comparable, we note that a multicenter study enrolling 416 patients with hand eczema recorded fewer positive patch test results in patients with a history of atopic dermatitis than in patients with no such history (55% vs 66%, respectively; P=.052).4 Other authors have found that the differences are not significant.13,16 Publications on contact eczema at any location, on the other hand, report a clear relation between atopic dermatitis and higher or lower frequencies of positive patch test results.14,20 These discrepancies may be attributable more to selection bias at the moment of deciding to order patch tests or not in atopic patients rather than to an influence of atopic dermatitis on sensitization in these patients.

Allergic contact dermatitis is seen largely in women, who also have significantly more positive patch test results than men both in contact eczema and hand eczema,4,13,20 consistent with our findings. Patch test positivity of current relevance was inversely associated with male sex in our study.

Although a diagnosis of chronic eczema is often the main indication for doing patch tests,1–6 we found no studies analyzing this aspect. In our study, lesion onset more than 6months earlier was inversely associated with finding a positive reaction to a currently relevant allergen, although the association was not confirmed in the multivariate analysis. The small size of our patient sample prevented us from dividing it into 2 groups according to time since lesion onset. A larger patient cohort that could be subdivided might suggest a cut-point for an association between time since onset and test positivity. Even considering the limitations of our study, however, we have shown that there are other clinical factors that are better indicators than chronicity for ordering patch tests.

In spite of limitations, our study offers a mathematical approach to prediction that may be very useful for decision-making in our clinical management of contact dermatitis. In our setting, all hairdresser–aestheticians who consult us for hand dermatitis should have patch tests performed regardless of how long lesions have been present. Our model suggests that a diagnosis of chronic eczema is not a necessary indication for patch testing. Rather, the dermatologist should consider the possibility that a patient might have been exposed to specific allergens, such as would be the case for hairdressers–aestheticians in contact with acrylates.

The authors declare that they have no conflicts of interest.