Promising Online Tool for Headache Epidemiology : The PRILEVEL Pilot Study

Objectives: The aim of this study was to develop and test an online questionnaire which would be used in headache epidemiological research in Estonia. Methods: An online questionnaire consisting of 14 headachetargeting questions was compiled. A specifically designed algorithm using the ICHD-3 beta criteria provided the headache diagnosis. All patients aged 18-65 years who had received a definite headache diagnosis by a specialist at Tartu University Hospital’s Headache Clinic from February 2014 to March 2015 were invited to complete the questionnaire. The diagnoses given by the headache specialists were compared to the diagnoses proposed by the algorithm to assess its sensitivity and specificity as well as its positive and negative predictive values. Results: The specificities and sensitivities of the main diagnostic groups were as follows: migraines 0.97 and 0.56, tension-type headaches 0.92 and 0.52, trigeminal autonomic cephalalgias 1 and 0.5, and other primary headaches 0.98 and 0.5 respectively. Incorporating ICHD-3 beta probable criteria in addition to the definite criteria for the migraine and tension-type headache diagnostic groups did not decrease the specificities markedly (0.9 and 0.92 respectively) but the sensitivity increased considerably (to 0.8 and 0.6 respectively). Conclusions: The PRILEVEL questionnaire and algorithm have very high specificity. Strictly applying the ICHD-3 beta definite criteria within the epidemiologic studies can lead to an underestimation of true prevalence values of primary headache disorders, and probable criteria should be included to increase sensitivity and decrease the influence of a recall bias.


Introduction
The results of epidemiological studies in the field of primary headaches vary to some extent [1,2].The prevalence studies that have been carried out so far do not cover all the world's countries or regions [1].It has been indicated that in Europe 50% of the population have an active headache disorder, whereas 15% suffer from migraine, 4% have chronic headache, and 1-2% have medication overuse headache [2].The overall prevalence of the tension-type headache in Europe has been estimated to be 62.6% [2].Cluster headache has a lifetime prevalence of 0.1-0.3%[3].Epidemiologic data on other primary headache disorders is scarce -Schwaiger et al. have reported the one-year prevalence of all other primary headache disorders to be ≤ 1.2%, whereas the prevalence for different diagnoses was as follows: primary stabbing headache 1.2%, primary cough headache 0.2%, primary exertional headache 0.2%, primary hypnic headache 0.2%, primary headache associated with sexual activity < 0.2%, primary thunderclap headache < 0.2%, and new daily-persistent headache < 0.2% [4].There are no specific data on the true prevalence of nummular headache; however, more than 250 cases have been described and in one hospital series the incidence was 6.4/100,000/year [5,6].Previously used methods of data collection in the majority of the epidemiological studies of headache prevalence have been either personal interview, telephone interview, or self-applied questionnaire [2].A novel online approach has been reported in two studies.A web-based survey of exercise-related headache was described by Ende-Kastelijn et al. in 2012, where all the participants in a tough cycling event in Holland were invited to fill in an online questionnaire in order to establish the prevalence of primary exertional headache among an athletic population [7].Wilbrink et al. have validated two stepwise webbased questionnaires (LUCA and QATCH) for diagnosing cluster headaches in order to carry out large-scale epidemiologic studies in Holland, and showed that such an approach is accurate and reliable, despite the relatively low sensitivity values of 53.8-57.2%[8].
The prevalence of headache disorders in Estonia has never been investigated.Estonia is one of the leading countries in the world regarding the usage of internet and web-based solutions per household -being 80% among the population of 16-74-yearolds in 2014 [9].Furthermore, the availability of internet and e-solutions covers the whole country.The coexistence of these two factors sets up potentially ideal conditions for using e-technology in performing representative epidemiological studies.
The aim of the present pilot study was to develop and test an online questionnaire which would be used in a nation-wide epidemiological study to estimate the one-year prevalence of migraines, tension-type headaches, trigeminal autonomic cephalalgias, other primary headaches, trigeminal neuralgia/ neuropathy, headache attributed to trauma to the head, and medication overuse headache in the general population in Estonia.

Methods
The study was approved by the Research Ethics Committee of Tartu University (permission no.242T-11).

The Questionnaire
An online questionnaire was compiled in Estonian and named PRILEVEL (acronym from Estonian PRImaarsete peavalude LEVimus ja ELukvaliteet, "the prevalence and quality of life with primary headaches").It had two parts.The first was designed to collect demographic data (age, height, weight, education, and residence in an urban vs. a rural area) and the lifestyle-related possible headache risk-factors (physical activity, smoking, and coffee and alcohol consumption) of participants.The last question of the first part was the screening question about the presence of headaches: "During the last year have you had repeated headaches not caused by an acute infection, a medication's side effects, medical procedures, or consumption of toxic substances including alcohol?"If a participant answered "yes", he/she was introduced to the second (diagnostic) part of the questionnaire.It consisted of 14 questions targeting different aspects of a person's headache (localization, laterality, character, intensity, preceding and accompanying symptoms, duration, frequency, response to indomethacin, association with certain situations/activities, precipitating factors, drug consumption, and history of head trauma).Based on the answers to these questions, a specifically designed sophisticated algorithm provided the diagnosis, whenever possible.The algorithm strictly used The International Classification of Headache Disorders, 3rd edition (beta version) (ICHD-3 beta) criteria to recognize the following entities: episodic migraine without aura, episodic migraine with aura, chronic migraine, tension-type headache (episodic/chronic), cluster headache (episodic/chronic), paroxysmal hemicranias (episodic/ chronic), short-lasting unilateral neuralgiform headache attacks (episodic/chronic), primary cough headache, primary exercise headache, primary headache associated with sexual activity, cold-stimulus headache, primary stabbing headache, nummular headache, hypnic headache, new daily persistent headache, headache attributed to trauma, trigeminal neuralgia, trigeminal neuropathy, and medication overuse headache [10].The algorithm always applied the ICHD-3 beta criteria for definite diagnoses except for two instances considering the diagnosis of migraine.Firstly, the duration of a migraine attack was allowed to last from 4 hours to 7 days in order to include patients who had attacks typical of migraine (4 h to 72 h) but who also had experienced status migrainosus.Secondly, the diagnostic criteria for definite migraine require that a person must have at least one of the followingnausea and/or vomiting or photophobia and phonophobia [10].The algorithm was allowed to give the diagnosis of migraine even if only photophobia or phonophobia was present if other migraine criteria were fulfilled.
The algorithm was not designed to recognize primary thunderclap headache, because this diagnosis requires an extensive diagnostic workup in order to exclude secondary causes, or external pressure headache, because the online approach is judged to be not sensitive enough to prove causality.
The algorithm was designed to propose only one diagnosis per each completed questionnaire except for the instance when medication overuse headache could be identified, which was then proposed additionally to the primary headache diagnosis.If the participant had more than one type of headache, he/she was asked to fill in the questionnaire again for each type of headache experienced.Hence, multiple diagnoses were allowed to avoid overlooking any co-morbid headaches.
The questionnaire was accessible via Tartu University Hospital's patient website, which is a highly secure website and requires digital identification upon entering.

Subjects
All persons aged 18-65 years who had received a definite headache diagnosis by a headache specialist at Tartu University Hospital's Headache Clinic from February 2014 to March 2015 were invited to complete the questionnaire.Subjects were informed of the questionnaire and the purpose of the study by telephone.Consenting patients were sent an e-mail which contained a direct link to the Tartu University Hospital's patient website, a short description of the questionnaire, and guidelines for how to fill it in.
After filling in the questionnaire and before saving the data, patients were once more informed of the purpose of the study and also that by pushing the "Complete" button, they confirmed their informed consent to participate in the study.

Statistical analysis
The diagnoses given by the headache specialists were compared to the diagnoses proposed by the algorithm.Sensitivity, specificity, and (prevalence-dependent) positive and negative predictive values of the algorithm were assessed in the patient cohort for each headache diagnosis (ICHD-3 beta second-digit-level diagnosis accuracy by the algorithm was required in order to count the diagnosis as correct) as well as for the larger groups of diagnoses in the study sample (in this case first-digit-level diagnosis accuracy was required to count the diagnosis as correct).After receiving the results of the statistical analysis of the named parameters of the algorithm using ICHD-3 beta definite criteria, the same parameters were recalculated with added ICHD-3 beta probable diagnostic criteria for the most prevalent groups of diagnoses -namely migraine and tension-type headache.In this case first-digit-level diagnosis accuracy was required to count the diagnosis as correct.

Enrolment of Patients
A total of 511 patients from Tartu University Hospital's Headache Clinic were included in the recruitment process, which started in January 2015 and was completed in March 2015.One hundred and forty-five participants were included for final analysis, representing a response rate of 28% (Figure 1).Demographic characteristics of the final study population are presented in Table 1.

Testing and Statistical Analysis
Using the headache specialist's diagnosis as the gold standard, the sensitivity and specificity as well as positive and negative predictive values of the algorithm using ICHD-3 beta definite criteria for each diagnosis were calculated and are depicted in Table 2.
After additionally applying the ICHD-3 beta probable criteria, the specificity values did not decrease markedly but the sensitivity increased considerably for the most prevalent headache diagnosis groups -the sensitivity for all migraines increased from 0.56 to 0.80 and for all tension-type headaches from 0.52 to 0.60; the specificity values remained 0.90 and 0.92 respectively (Table 3).

Discussion
To our best knowledge, this is the first pilot study applying ICHD-3 beta criteria to diagnose such a wide spectrum of headache disorders by using a single, simple web-based questionnaire and a digital algorithm.It must be stressed that the main objective of this study was to test the developed online questionnaire with its algorithm before using it in further epidemiological research.It must be acknowledged that such an approach is justifiable only within the appropriate setting, as it is in Estonia, where the availability of web-based solutions and their real usage is sufficiently high and has already become a rather natural part of everyday life in the country.The latter means that the vast majority of the general population (especially those under investigation -that is, 18-65-year-olds) are actively and comfortably in touch with e-solutions on an everyday basis.
One of the main strengths of this study is its simplicity, which should lead to an optimization of usage of limited resources.When compared to the methodology of previously published epidemiological studies, this novel approach could provide not only data representative of a general population, but also considerably increase the power of studies by including the largest possible percentage of the general population.Realistically speaking, the latter cannot be achieved when using other approaches such as the door-to-door or telephone interviews.In addition, it is also important that both the rural and urban population are sufficiently covered: in our study sample, 77.2% of the participants lived in urban areas and 22.8% in rural areas.According to the last population and housing census in Estonia in 2011, 69% of residents aged 18-64 lived in urban settlements [11], which is slightly less than in our study (p = 0.03).On the other hand, this is expected because the urban population is more active in using web-based solutions.
Although the usage of the internet in Estonia is exceptionally high, it is not used by everyone; approximately 80% of the adult  population uses it.Hence, there is a smaller part of the general population that would potentially not be covered with this methodological approach.This issue may be considered a source of bias and should be investigated further.This is also the reason why this novel online-approach will be compared to the classical doorto-door method in a random sample within the upcoming nationwide epidemiological study.Still, only 8.8% of the initial study sample could not participate due to participants not using digital identification methods or their lack of access to the internet (7.4% and 1.4% respectively).
The questionnaire itself is relatively simple and easy to complete, which is supported by the finding that only 4.1% of the participants completed the questionnaire incorrectly, either by not answering all the compulsory questions or by giving illogical answers.Another strength of the online questionnaire is that since digital identification is required by the participant, it allows the collection of personalized data, which in turn enables the elimination of multiple data entries and the acquisition of accurate prevalence rates of primary headache disorders.
The questionnaire and the algorithm were first strictly based on the ICHD-3 beta definite criteria -that is why the questionnaire has very high specificities.Sensitivities of some of the headache diagnoses in the study sample, however, turned out to be lower.This factor concerns, in particular, episodic migraine with aura, chronic migraine, primary stabbing headache, and nummular headache.One of the reasons the algorithm did not recognize the precise diagnosis in the case of migraines was related to the participants reporting their headache duration without treatment to be less than 4 which automatically excluded the definite migraine diagnosis, according to ICHD-3 beta.In some cases, the migraine diagnosis was not detected, because of the recall bias, when participants did not report any accompanying symptoms, although these were clearly stated in their headache specialist history.In some instances, the algorithm did not propose a definite diagnosis, because the case itself was atypical and thus unrecognizable by the strict criteria, although the physician's diagnosis was established as a definite primary headache.Except for the aforementioned values the sensitivities for other diagnoses were 0.5 or above.Similar sensitivity values (53.8-57.2%)were also achieved by Wilbrink et al. for the LUCA and QATCH online questionnaires [8].Based on their findings, the authors concluded that such an approach is acceptably accurate and reliable for epidemiological studies [8].In addition, there are two important points that must be noted when interpreting the calculated sensitivity values.First, the present statistical methodology for calculating sensitivity values requires an exact knowledge of the prevalence rates of the disorder(s) in the particular settings where the study is being performed.Since Estonia has no such data, the prevalence rates needed for the analysis were adopted from studies performed elsewhere.The latter ones, however, vary to a considerable extent (but not dramatically).Hence, based just on the adopted prevalence rates, which were selected by authors prognostically based on their best informed knowledge, the real sensitivity values might be somewhat different from the presented ones.Secondly, it has to be acknowledged that since the values of the sensitivities and specificities have been calculated on a cohort of "pre-educated" patients, as they have already been to a headache specialist, these estimates cannot be directly applied to the general population.Both statements require us to avoid over emphasizing the rather low sensitivity values; due to the unavoidable limitations of the present setting, the pilot-study was performed as well as it could have been.Furthermore, including probable diagnoses in the algorithm increased sensitivity substantially (in the case of migraine, from 0.56 to 0.80) by detecting those cases that did not fulfill the definite diagnostic criteria but in real life lead to the clinical diagnosis by headache specialists and/or other physicians.It also addresses another important issue -minimizing the influence of a recall bias (patients forgetting to report one or several headache descriptors) on the results, which in turn brings the prevalence rates closer to real ones.Based on our findings, it can be suggested that strictly applying the ICHD-3 beta definite criteria within the epidemiologic studies can lead to an underestimation of true prevalence values of primary headache disorders.
The quality control of the clinical diagnosis in our study can be rated as at least very good, since all subjects were personally consulted and a headache disorder diagnosis was made by a Tartu University Hospital's Headache Clinic headache specialist.Despite all the strengths of the present pilot study, some limitations must be mentioned.The response rate of the study is not very high (28%).Nevertheless, it is considered acceptable.For instance, when publishing their results on the validation of the QATCH web-based questionnaire, Wilbrink et al. had the same parameter at 20% [8].A rather low response rate was one of the factors contributing to the fact that our study sample was relatively small and all the diagnoses that the algorithm was designed for were not covered.For example, there were no subjects with rather rare headache disorders such as chronic cluster headache, paroxysmal hemicrania, short-lasting unilateral neuralgiform headache attacks, primary cough headache, primary headache associated with sexual activity, cold-stimulus headache, or hypnic headache.For some diagnoses, namely episodic cluster headache, primary exercise headache, primary stabbing headache, nummular headache, and trigeminal neuralgia, there were only a few subjects, so the conclusiveness for these entities is not strong.This is also the main reason why the sensitivities for primary stabbing headache and nummular headache were extremely low.
In conclusion, the PRILEVEL online questionnaire and algorithm have very high specificity.Adding the ICHD-3 beta probable criteria to the algorithm increases sensitivity values to an acceptable level for usage of the tool in a nation-wide epidemiological study and is highly recommended for other studies using a similar approach elsewhere.

Figure 1 :
Figure 1: Flowchart of the recruitment of the subjects.

Table 1 :
Demographic characteristics of the study participants.

Table 2 :
Specificity, sensitivity, positive (PPV), and negative predictive values (NPV) of the PRILEVEL algorithm for the diagnoses in the study sample.

Table 3 :
Specificity, sensitivity, positive (PPV) and negative predictive values (NPV) of the PRILEVEL algorithm for the migraine and tension-type headache diagnoses after incorporating the probable criteria.