To identify a set of clinical parameters that can predict the probability of carrying mutations in one of the genes associated with hereditary autoinflammatory syndromes.
To identify a set of clinical parameters that can predict the probability of carrying mutations in one of the genes associated with hereditary autoinflammatory syndromes.
A total of 228 consecutive patients with a clinical history of periodic fever were screened for mutations in the MVK, TNFRSF1A, and MEFV genes, and detailed clinical information was collected. A diagnostic score was formulated based on univariate and multivariate analyses in genetically positive and negative patients (training set). The diagnostic score was validated in an independent set of 77 patients (validation set).
Young age at onset (odds ratio [OR] 0.94, P = 0.003), positive family history of periodic fever (OR 4.1, P = 0.039), thoracic pain (OR 4.6, P = 0.05), abdominal pain (OR 33.1, P < 0.001), diarrhea (OR 3.3, P = 0.028), and oral aphthosis (OR 0.2, P = 0.007) were found to be independently correlated with a positive genetic test result. These variables were combined in a linear score whose ability to predict a positive result on genetic testing was validated in an independent data set. In this latter set, the diagnostic score revealed high sensitivity (82%) and specificity (72%) for discriminating patients who were genetically positive from those who were negative. In patients with a high probability of having a positive result on genetic testing, a regression tree analysis provided the most reasonable order in which the genes should be screened.
The proposed approach in patients with periodic fever will increase the probability of obtaining positive results on genetic testing, with good specificity and sensitivity. Our results further help to optimize the molecular analysis by suggesting the order in which the genes should be screened.
Familial Mediterranean fever (FMF; MIM no. #249100), tumor necrosis factor receptor–associated periodic fever syndrome (TRAPS; MIM no. #142680), and mevalonate kinase deficiency (MKD; MIM no. #260920) are hereditary diseases caused by mutations of genes involved in the regulation or activation of the inflammatory response (1). These conditions are characterized by recurrent episodes of fever associated with a number of manifestations: rash, serositis, lymphadenopathy, and arthralgias/arthritis. Disease onset is commonly observed during the pediatric years, and a delay in the molecular diagnosis frequently occurs. Systemic reactive AA amyloidosis may be a severe long-term complication of these conditions.
FMF is caused by recessive mutations in the MEFV gene, which encodes a protein called pyrin (2, 3). TRAPS is due to dominant mutations in TNFRSF1A, the TNF receptor superfamily 1A gene (4). MKD is caused by recessive mutations in MVK, the mevalonate kinase gene (5, 6).
The molecular diagnosis of these diseases has low efficiency and, consequently, elevated costs, for 2 main reasons. First, in patients with clinical manifestations consistent with an autoinflammatory syndrome, the rate of detection of mutations in any of these 3 genes is usually <20% (7–9). Second, the clinical manifestations of the 3 disorders are largely overlapping. Although ethnicity, family history, duration of the attacks of fever, and the presence of rare, peculiar clinical manifestations may be helpful, it is often difficult to decide which gene should be screened for first based on the clinical features. This is particularly true in pediatric patients, since there is the rather frequent occurrence of another clinical entity characterized by recurrent attacks of fever, which is known as periodic fever, aphthosis, pharyngitis, and adenitis (PFAPA) syndrome (10, 11). To date, this syndrome does not have a documented genetic basis, and it usually undergoes spontaneous resolution of the episodes of fever a few years after symptom onset.
In the present study, we developed and validated a set of variables that predict the risk that a given patient carries a mutation in 1 of the 3 genes known to be associated with recurrent fever. The diagnostic flow chart we propose may improve the genetic diagnostic evaluation of children with periodic fevers of unknown origin and reduce the related costs of analyses.
The study was conducted using data from 2 patient samples. From the initial screening group, a training sample was derived and was used to generate hypotheses (i.e., to create the diagnostic model). A validation sample was then used to test the performance of the diagnostic model in an independent population of patients.
Starting from 2002, a nationwide laboratory facility for the genetic diagnosis of autoinflammatory disorders in children was established at the Gaslini Institute. Detailed information about each patient's family history, personal history, and clinical manifestations were collected (data available on the authors' Web site at http://www.printo.it/periodicfever).
Criteria for inclusion in the screening group consisted of the following: 1) periodic attacks of fever (<38°C) of unknown origin, with fever-free and symptom-free intervals characterized by normal levels of acute-phase reactants, and 2) at least 2 of the following symptoms present during the attacks of fever: lymphadenopathy, splenomegaly, gastrointestinal manifestations, chest pain, mucocutaneous manifestations, musculoskeletal manifestations (data available on the authors' Web site at http://www.printo.it/periodicfever).
Up to July 2005, specimens from 1,347 patients were received for molecular diagnosis. Of these, 119 were excluded from the present study because of 1 or more of the following reasons: 1) did not meet the inclusion criteria, 2) lack of complete clinical information, and 3) patient was recruited and screened for suspected cryopyrin-related disorders (12, 13). Thus, the results of a genetic analysis of all 3 genes involved in periodic fevers (MEFV, TNFRSF1A, and MVK) and complete clinical information were available in 228 of the 347 patients.
For autosomal-recessive diseases (MKD and FMF), only homozygous or compound heterozygous patients were selected as being genetically positive (8). Conversely, patients carrying heterozygous TNFRSF1A missense mutations known to be associated with a severe phenotype (14–16) were distinguished from patients carrying low-penetrance, mild mutations, such as R92Q and P46L (14, 17). In fact, the actual pathogenic relevance of these latter mutations is still a matter of debate (14, 18). Thus, the following 6 subgroups were arbitrarily identified among the 228 patients who were screened: MKD patients, severe TRAPS patients, mild TRAPS patients, FMF patients, MEFV heterozygous patients, and negative patients.
The training set of patients consisted of those who were classified as positive if the genetic testing yielded a definite genetic diagnosis (either MKD, severe TRAPS, or FMF) and those who were classified as negative if the genetic testing showed no mutation of the 3 genes involved in the hereditary autoinflammatory disorders under study. Even if some of the patients carried low-penetrance TNFRSF1A or a single MEFV mutation and had clinical manifestations consistent with the respective disease, we decided to exclude them from the training set.
An additional group of patients was recruited in whom the diagnostic score that was developed in the training set of patients could be validated. This validation set included children who were diagnosed at Giannina Gaslini Institute after July 2005 and children who had already been diagnosed at other European centers. With regard to these latter patients, their treating physicians, who were unaware of the results obtained in the training set of patients, were asked to retrospectively complete the same information forms as were used for the Italian population. Patients in the validation set were classified according to the same criteria that were applied to those in the training set.
The study was approved by the Ethical Board of Giannina Gaslini Institute. Written informed consent was obtained from each enrolled patient, the patient's parents, or the patient's legal guardian.
The extracellular region of the p55 TNF receptor (from exon 1 to exon 6) of the TNFRSF1A gene, the 10 coding exons (from exon 2 to exon 11) of the MVK gene, and exons 2, 3, 5, and 10 of the MEFV gene were analyzed as previously described (16, 19). Exons 2, 3, 5, and 10 of the MEFV gene were amplified and directly sequenced, using polymerase chain reaction primers (available upon request from the corresponding author) under standard conditions (16, 19). In patients heterozygous for MEFV mutations, rare mutations of exons 1, 4, 6, 7, 8, and 9 were tested (results not shown).
A diagnostic score was constructed with the primary aim of discriminating patients who were likely to have positive results on genetic testing from those who were likely to have negative results, and therefore need not undergo genetic testing. As a secondary and exploratory end point, we attempted to assess the most reasonable order in which the genetic tests should be performed in patients who, after the first step, were determined to be at high risk. The diagnostic score was constructed and validated through the following 4 steps.
Patients in the training set were classified as having positive or negative results on genetic testing, according to the criteria outlined above. A univariate logistic analysis was performed to identify the symptoms and clinical variables that were significantly associated with a positive genetic test result. Only symptoms that were present in at least 5% of the patients were studied. Variables yielding a P value of less than 0.20 were entered into a multivariate logistic analysis in which the positivity of the genetic test was the dependent variable. Once this model was determined, a linear diagnostic score was calculated using the variables included in the model (coded so as to give the best fit) and the corresponding estimated coefficients, according to the following formula:
where βi represents the coefficients estimated by the multivariate model and vari represents the variables included in the model. A receiver operating characteristic (ROC) curve with area under the curve (AUC) was used to evaluate the performance of the diagnostic score in the training set. In particular, the cutoff value of the diagnostic score that was chosen to classify patients as low risk and high risk was defined as the point that gave an optimal level of sensitivity and specificity.
In the validation sample, the performance of the diagnostic score, which was calculated using the same coefficients and the same cutoff value estimated with the training set, was assessed by testing its ability to discriminate patients who were genetically positive from those who were genetically negative. A ROC curve with AUC was used to assess the performance of the model. The entire procedure for calculating the score for each of the individual patients and the technical details of the procedure are outlined on the authors' Web site at http://www.printo.it/periodicfever.
In patients from the training set who were classified by the diagnostic score as being positive, a classification was performed using a regression tree analysis. The aim of this analysis was to determine which genetic test should be performed first, by predicting the probability of being in a particular class (i.e., disease group: MKD, TRAPS, or FMF) based on the values for the independent variables (symptoms). Only symptoms that were significantly correlated at a 1% significance level with a specific disease category in the univariate analysis (by chi-square test) were included in the regression tree analysis.
The rules developed for assigning patients to 1 of the 3 disease categories estimated in the training set were applied to patients in the validation set who, at step 2, were predicted to be genetically positive, and the proportion of patients who were correctly classified was assessed in order to verify the validity of the results obtained in the training set. Since these last steps were based on a small number of patients, the results obtained must be considered preliminary, requiring further validation in larger groups of patients.
The clinical and genetic features of the entire group of 228 patients are reported in Table 1, as well as on the authors' Web site at http://www.printo.it/periodicfever. At the time of genetic testing, the majority of the patients (204 [89.4%]) were under the age of 18 years (mean 8.9 years [range 0.5–65 years]). The mean age at disease onset was 4.3 years (range 0.1–17.8 years).
|Symptom||Entire group of patients screened (n = 228)||Validation set of patients (n = 77)|
|MKD (n = 18)||Severe TRAPS (n = 7)||FMF (n = 12)||Mild TRAPS (n = 15)||Heterozygous MEFV (n = 40)||Negative (n = 136)||MKD (n = 13)||Severe TRAPS (n = 5)||FMF (n = 13)||Mild TRAPS (n = 3)||Heterozygous MEFV (n = 3)||Negative (n = 40)|
|Positive family history of periodic fever||17||86||17||7||18||8||0||80||31||0||0||8|
|Cervical lymph node enlargement||94||43||42||60||58||63||92||20||20||67||67||78|
|Pain in cervical lymph nodes||72||14||33||33||35||27||85||0||0||33||0||38|
|Age at onset, mean ± SD months||10.4 ± 8.3||17.9 ± 17.1||16.6 ± 11.2||58.1 ± 64.4||29.6 ± 44.6||49.5 ± 58.2||12.0 ± 6.3||45.0 ± 38.0||35.1 ± 28.9||48.0 ± 24.0||24.3 ± 20.2||42.9 ± 56.8|
|Duration of fever, mean ± SD days||4.3 ± 1.4||15.3 ± 7.8||3.0 ± 1.9||4.7 ± 3.7||5.9 ± 9.1||5.4 ± 7.5||4.2 ± 1.0||14.4 ± 9.3||2.2 ± 0.9||9.0 ± 5.2||4.0 ± 2.6||5.7 ± 3.8|
Eighteen patients were identified as being either homozygous or compound heterozygous for the MVK gene (Table 1) (19). None of the MKD patients had clinical manifestations that were consistent with mevalonic aciduria.
Twenty-two patients were identified as having heterozygous mutations of the TNFRSF1A gene. Six patients who had severe missense substitutions in the TNFRSF1A gene and 1 patient who had an in-frame interstitial deletion of 27 nucleotides were considered to have severe TRAPS (16). In addition, we identified 13 low-penetrance R92Q mutations, 1 D12E missense mutation, and 1 c.194-14G>A splice mutation (16) (data available on the authors' Web site at http://www.printo.it/periodicfever). At the time statistical analyses were conducted, these latter 2 patients experienced an almost complete spontaneous resolution of their clinical manifestations and were therefore arbitrarily included in the group with mild TRAPS.
Fifty-two patients carried at least 1 mutation in the MEFV gene. Twelve of these patients were either homozygotes or compound heterozygotes, and the remaining 40 patients were heterozygotes (data available on the authors' Web site at http://www.printo.it/periodicfever).
Of these 92 patients with positive findings on genetic testing, 3 also displayed mutations of another gene examined in this study. Two of these 3 patients, who were compound heterozygotes for the MVK gene, also carried the TNFRSF1A R92Q mutation and were classified as having MKD (16). The third patient, who was homozygous for the M680I mutation of the MEFV gene, also carried the TNFRSF1A D12E mutation and was classified as having FMF.
A total of 136 patients with periodic or recurrent fevers did not display any mutation of the 3 genes we examined (Table 1).
After the screening procedure was completed, 173 of the 228 patients were included in the training set. This group consisted of the 18 patients with MKD, the 7 patients with severe TRAPS, the 12 patients with FMF, and the 136 patients who were negative for the 3 genes we examined.
The 173 patients who were included in the training set were grouped according to positive (37 patients) and negative (136 patients) results on genetic testing.
Table 2 shows the findings of the univariate logistic regression analysis (with the positive results on genetic testing as the dependent variable) and the variables retained in the final multivariate model. All symptoms were studied in univariate analysis as binary variables (symptom present/absent); age and duration of fever were studied as continuous covariates. Young age at disease onset, positive family history of periodic fever, thoracic pain, abdominal pain, diarrhea, and oral aphthosis were the variables that independently predicted the probability of having a positive genetic test result (Table 2) and were included in the final model.
|Independent variable||Univariate analysis||Multivariate analysis (final model)|
|OR (95% CI)||P||OR (95% CI)||P|
|Age at onset, months||0.96 (0.94–0.99)||0.005||0.94 (0.91–0.98)||0.003|
|Positive family history of periodic fever||4.6 (1.8–11.5)||0.001||4.1 (1.1–16.0)||0.039|
|Days of fever||1.01 (0.97–1.05)||0.73|
|Oral aphthosis||0.5 (0.2–1.1)||0.09||0.2 (0.1–0.7)||0.007|
|Erythematous pharyngitis||1.1 (0.5–2.4)||0.74|
|Exudative pharyngitis||0.6 (0.3–1.3)||0.19|
|Periorbital edema||2.7 (0.7–10.3)||0.13|
|Cervical lymph node enlargement||1.4 (0.6–2.9)||0.44|
|Pain in cervical lymph nodes||2.7 (1.3–5.7)||0.009|
|Thoracic pain||3.9 (1.3–11.6)||0.01||4.6 (1.0–22.5)||0.05|
|Abdominal pain||21.3 (4.9–92.1)||<0.001||33.1 (6.1–178.5)||<0.001|
|Diarrhea||5.0 (2.3–10.8)||<0.001||3.3 (1.1–9.8)||0.028|
The variables retained in the final model were re-coded in order to achieve the best fit. The diagnostic score was calculated using the linear combination of these variables, which were weighted according to the coefficients (log of the odds ratios) estimated by the logistic model (Table 3). Some numerical examples of the diagnostic score calculation are reported on the author's Web site at http://www.printo.it/periodicfever.
|Age at onset||Months||−0.067|
|Abdominal pain||Never = 0|
|Sometimes or often = 2||1.494|
|Always = 3|
|Aphthosis||Never = 0|
|Sometimes or often = 1||−1.504|
|Always = 2|
|Thoracic pain||Absent = 0||1.958|
|Present = 1|
|Diarrhea||Never = 0|
|Sometimes = 1||0.901|
|Often = 2|
|Always = 3|
|Family history||Negative = 0||1.503|
|Positive = 1|
The quartiles of the scores were −2.30, −0.22 (median value), and +2.53. The ROC curve for the diagnostic score is illustrated on the author's Web site at http://www.printo.it/periodicfever. The AUC was 0.95, indicating a high discriminative ability of the diagnostic score in the training set. The cutoff value of the diagnostic score for classifying patients as low risk and high risk was chosen from the ROC curve. Patients were classified as high risk if their diagnostic score was >1.32; otherwise, they were classified as low risk. This cutoff value guaranteed high sensitivity and good specificity, as shown in Table 4. A total of 35 (95%) of the 37 patients with a positive genetic diagnosis were correctly classified as being positive according to the diagnostic score (sensitivity 95%), and only 1 patient with severe TRAPS and 1 patient with FMF according to the genetic diagnosis were classified as being negative according to the diagnostic score. Among the 136 patients with a negative genetic test result, 24 (18%) were erroneously classified as being positive, yielding a specificity of 82% for the diagnostic score.
|Diagnosis predicted by the diagnostic score||Total|
|TRAPS||6 (86)||1 (14)||7|
|FMF||11 (92)||1 (8)||12|
|Negative||24 (18)||112 (82)||136|
|MKD||12 (92)||1 (8)||13|
|TRAPS||4 (80)||1 (20)||5|
|FMF||11 (85)||2 (15)||13|
|Negative||12 (30)||28 (70)||40|
|Not defined†||1 (17)||5 (83)||6|
Fifty-five patients with an incomplete genotype (40 heterozygous for MEFV mutations and 15 with low-penetrance or uncertain TNFRSF1A mutations) were excluded from the training set. When these patients were evaluated according to the diagnostic score, 17 patients heterozygous for MEFV and 4 with mild TRAPS carrying the R92Q mutation had a score that was >1.32. The remaining 34 patients (61.8%) were classified as being at low risk of carrying a mutation in any of the 3 genes considered.
A total of 77 patients were enrolled in the validation set; 52 were screened at Gaslini Institute after July 2005 and 36 were from other European centers. Of these 77 patients, 31 had positive findings on genetic testing, 6 had an undefined diagnosis (3 with low-penetrance TNFRSF1A mutations and 3 heterozygous for MEFV), and 40 were negative (Table 1) (gene mutation data are available on the authors' Web site at http://www.printo.it/periodicfever). The mean age of these patients at the time of genetic testing was 9.3 years (range 2–26 years). The mean age at disease onset was 3.3 years (range 0.1–14 years).
Based on the patients' clinical features and on the parameters estimated in the training set (Table 3), the diagnostic score was calculated for each patient in the validation set. Patients were then classified as low risk or high risk according to the cutoff score that was defined in the training set. The ROC curve for the diagnostic score in the validation set (the 6 patients with an undefined genetic diagnosis were considered negative) lay below the ROC curve in the training set, indicating a decrease in the discriminant ability of the diagnostic score in an independent data set (graphic representation of the ROC curve is available on the authors' Web site at http://www.printo.it/periodicfever). However, the AUC was 0.85, indicating good accuracy in predicting whether patients were positive or negative (see graphic representation of the ROC curve, available on the authors' Web site at http://www.printo.it/periodicfever).
When classifying patients as positive or negative according to the same cutoff value applied to the training set (score of 1.32), 27 of the 31 patients with positive findings on genetic testing (sensitivity 87%) were correctly classified as being positive according to the diagnostic score (Table 4). One MKD patient, 1 TRAPS patient, and 2 FMF patients were erroneously classified as negative, whereas 13 of the 46 genetically negative patients were erroneously classified as positive (specificity 72%). Five of the 6 patients who were heterozygous for MEFV or had the low-penetrance R92Q and P46L TNFRSF1A mutations were classified as negative according to the diagnostic score.
Our final aim was to identify, for each patient with a high probability of being positive according to the diagnostic score, the clinical clues that would be able to orient the most reasonable order of the genes to be screened. To this end, all 37 patients in the training set who had a definite genetic diagnosis were studied with a regression tree classification analysis to find symptoms that would be able to predict the genetic diagnosis (MKD, TRAPS, or FMF). The signs or symptoms with a significantly different incidence (at a 1% significance level) in the 3 groups, as determined by univariate analysis, were exudative pharyngitis (P = 0.008), enlargement of the cervical lymph nodes (P = 0.001), cervical lymph node pain (P = 0.01), pleurisy (P = 0.005), vomiting (P = 0.005), splenomegaly (P = 0.003), and mean duration of episodes of fever (P < 0.001).
All of these symptoms were included in a regression tree model and selected by a stepwise procedure. The results of the regression tree analysis are shown in the boxed area of Figure 1. The mean duration of episodes of fever (P < 0.001), vomiting (P = 0.004), and splenomegaly (P = 0.05) were the symptoms that best differentiated the 3 disease groups.
These same rules shown in the boxed area of Figure 1 were applied to the patients in the validation set who were previously classified by the diagnostic score as being at high risk (40 patients) (Table 4). The results of the entire classification procedure in the validation set (diagnostic score plus regression tree analysis) are reported in Table 5. Of the 31 patients with positive genetic test results, 24 (77%) were correctly classified. Two FMF patients were erroneously classified as having MKD, and 1 TRAPS patient was erroneously classified as having FMF. Thus, the good performance in classifying the most probable genetic diagnosis obtained in the training set was confirmed in the validation set, providing reliable indications of the most appropriate genetic test to be performed.
|MKD||TRAPS||Mild TRAPS||FMF||Heterozygous for MEFV||Negative|
|Diagnosis predicted by the diagnostic score|
|MKD||12†||0||0||2 (NCC)||1 (NCC)||6 (FP)||21|
|FMF||0||1 (NCC)||0||9†||0||3 (FP)||13|
|Negative||1 (FN)||1 (FN)||3||2 (FN)||2||28†||37|
Clinical criteria for FMF (20) and PFAPA syndrome (11) have been established. We therefore investigated the sensitivity and specificity of these criteria in our population of children with periodic fever in the training set.
Of the 228 patients, 122 fulfilled the criteria. All 12 patients who were homozygous or compound heterozygous for MEFV gene mutations were positive for FMF according to the clinical criteria. However, a relevant number of MKD patients (14 of 18) and of genetically negative patients (66 of 136) also met these criteria for FMF, demonstrating that patients with a genetic diagnosis other than FMF can still satisfy the clinical criteria for FMF.
One hundred eleven of the 228 patients fulfilled the PFAPA criteria, among which were 61 of 136 genetically negative patients, as well as 20 of 37 patients with a definitive genetic diagnosis (15 MKD, 4 TRAPS, and 1 FMF). These criteria therefore do not represent a specific tool for selecting patients who have a high probability of having negative genetic test results.
In the present study, we sought to set up an evidence-based diagnostic score by which to sort patients with periodic fever according to their risk of having positive results on pertinent genetic tests. In particular, the present study allowed us to identify some clinical variables (positive family history of periodic fever, early age at disease onset, presence of abdominal and chest pain, diarrhea) that are highly associated with the probability of detecting a relevant mutation(s) in known periodic fever syndrome genes.
Our study was made possible by the availability of a large number of patients with periodic fever who were systematically evaluated for the MVK, MEFV, and TNFRSF1A genes. Other hereditary autoinflammatory diseases, such as those associated with mutations of the Cryopyrin and NOD2/CARD15 genes, were not included in the study because, unlike the 3 disorders we examined, these conditions are characterized by a chronic or subchronic disease course and distinctive and highly suggestive clinical features (i.e., early-onset urticarial rash, deafness, bone deformities, or chronic granulomatous polyarthritis) (13, 21). Although the vast majority of pediatric patients with CIAS1 mutations have a chronic or subchronic disease course (13, 22), it is noteworthy that among the cryopyrinopathies, the intermediate form (defined as Muckle-Wells syndrome) may sometime present as recurrent episodes of systemic inflammation, mainly characterized by low-grade fever (usually below 38°C) and urticarial rash, thus suggesting the differential diagnosis of hereditary periodic fevers.
Autoinflammatory diseases characterized by mutations in the genes for FMF, TRAPS, and MKD appeared to be homogeneously distributed in our cohort of patients, whereas in different geographic settings, ethnicity may markedly influence the frequency of each disease, particularly FMF (8, 9, 23). An unexpectedly high prevalence of MKD was observed in our pediatric Italian population, especially in comparison with the prevalence of FMF. With the exception of southern Italy (mainly Sicily and Calabria), where the frequency of FMF is higher than in any other part of the country, the entire prevalence of FMF was certainly lower than that observed in other Mediterranean countries and ethnicities (23), as has been observed in Western Caucasian populations (9). Moreover, while MKD is consistently characterized by a very early onset (during the first years of life), up to 40% of FMF patients present with disease onset during the second decade of life or later (23, 24), thus accounting for the relatively low frequency of FMF in our groups of study patients.
Our multivariate analysis of the training set of patients identified a number of clinical variables that were strongly related to the probability of a positive genetic test result and allowed the development of a diagnostic score based on these variables. The resulting diagnostic test (Table 4), once it was verified in the validation set of patients, revealed high sensitivity (87%) and good specificity (72%) for the identification of genetically positive and negative patients.
Although the positive (and negative) predictive value of the score depends on the prevalence of the gene mutations, the sensitivity and specificity of the score should remain the same in each population. However, an external validation of the diagnostic score is warranted, especially in populations of different ethnic and genetic backgrounds and in populations living in different environments that might possibly interact with the genotypes examined.
A second aim of the present study was to identify among patients with a high risk of having positive genetic test results the clinical features that confer the highest probability of having each of the 3 autoinflammatory diseases. As expected, the duration of episodes of fever was able to discriminate most of the patients with TRAPS and FMF. Since there is an overlap between MKD and FMF in terms of the number of days of fever, the presence of vomiting and splenomegaly turned out to be an additional good indicator of the presence of MKD. The application of the regression tree analysis to the validation set allowed the correct identification of the mutated gene in 24 of 27 genetically positive patients, showing very good accuracy (89%).
Based on our results, we propose that the flow chart shown in Figure 1 be used as a guideline for genetic testing in children with recurrent fever of unknown origin. This flow chart includes calculation of the diagnostic score (see additional illustrative materials available on the authors' Web site at http://www.printo.it/periodicfever) in each patient according to the presence and frequency of the clinical manifestations associated with that patient's episodes of fever. The 2-step approach was motivated by the primary need to identify patients who, based on their clinical profile, were likely to have negative results on genetic testing, and therefore need not undergo such analysis. In contrast, in patients who are identified as being at high risk of having positive results on genetic testing (i.e., those with a diagnostic score >1.32), the molecular analysis can be performed as indicated by the regression tree analysis. We suggest that patients with a risk lower than the threshold be followed up longitudinally to identify possible new clinical manifestations or possible spontaneous reductions in the frequency and severity of episodes of fever. It is noteworthy that the preliminary results of a followup study of genetically negative patients included in the present cohort showed that a relevant number of the patients experienced spontaneous resolution of their episodes of fever within 2 years of genetic testing (11). Moreover, longitudinal evaluation of patients who were excluded during our screening step may allow the detection of patients who are indeed positive for 1 of the 3 diseases we examined, as was observed in 4 patients who were included in the validation set.
In our experience, application of the flow chart to the validation set of patients would have allowed the correct identification of 24 of the 31 genetically positive patients and 28 of the 40 genetically negative patients, thus avoiding genetic screening in 28 patients for 3 genes and in 24 patients for 2 genes, resulting in a total of 132 genetic tests (and concomitant costs) that could have been avoided, at least at the first screening phase. This flow chart may be integrated with laboratory examinations, such as serum levels of IgD and urinary excretion of mevalonic acid during episodes of fever, in patients in whom a diagnosis of MKD is suspected, although the former test has a low specificity for the disease (19, 25), whereas the latter test is available only in a few specialized laboratories.
The majority of children with periodic fevers do not carry a mutation in the known periodic fever syndrome genes. To complicate matters, the most common periodic fever syndrome, PFAPA (10, 11), a benign, self-limited syndrome, has a clinical phenotype that overlaps with that of some monogenic autoinflammatory diseases (mainly MKD). In the present study, we have shown that the currently available diagnostic criteria for PFAPA syndrome are not able to distinguish genetically positive patients from genetically negative patients. Similarly, when we assessed the performance of the clinical criteria for FMF (20) in our population of Italian children with periodic fever, we found that although all FMF patients who were homozygous or compound heterozygous for MEFV gene mutations were also positive according to the clinical criteria, a relevant number of MKD patients and genetically negative patients also met the criteria for FMF.
Federici et al (8) recently performed a similar study in a large group of adult patients who were screened for the MEFV, MVK, and TNFRSF1A genes. According to their study, meeting the FMF criteria and being of Mediterranean origin should be considered clear indications for undergoing molecular analysis of the MEFV gene. In the absence of these features or in the presence of a noninformative result on genetic testing for MEVF, the choice of the next gene to be screened should be made on the basis of expert advice (8). Since almost all of our patients were of Italian origin, the multivariate analysis was not able to evaluate the ethnic origin of the patients among the possible discriminating variables. However, it is conceivable that this variable should be included in our diagnostic flow chart (Figure 1), especially in populations characterized by a high prevalence of mutations of a given gene, as is the case for the MEFV gene in Jewish, Arabian, Armenian, and Turkish populations (8).
In conclusion, we propose an evidence-based guideline that could help general pediatricians and clinical geneticists in the diagnostic evaluation of children with periodic fever. Further evaluation in longitudinal studies and in different populations will help to sharpen the sensitivity and specificity of the proposed diagnostic score and flow chart.
Dr. Gattorno had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Gattorno, Sormani, Martini.
Acquisition of data. Gattorno, Sormani, D'Osualdo, Caroli, Federici, Cecconi, Pelagatti, Solari, Meini, Zulian, Obici, Breda, Martino, Tommasini, Bossi, Govers, Touitou, Woo, Frenkel, Koné-Paut, Baldi.
Analysis and interpretation of data. Gattorno, Sormani, Pelagatti, Cecconi, Baldi, Ceccherini, Martini.
Manuscript preparation. Gattorno, Zulian, Touitou, Woo, Frenkel, Koné-Paut, Ceccherini, Martini.
Statistical analysis. Sormani.