Prevalence, aetiologies and prognosis of the symptom dizziness in primary care – a systematic review

We performed a systematic review including symptom evaluating studies about dizziness in a primary care setting. Symptom evaluating studies are defined as studies that examine patients presenting with a defined symptom in a given health care setting. These studies seek to investigate prevalence/incidence, underlying aetiologies and prognosis for patients presenting with this symptom [7].

We included all original research articles evaluating the symptoms “dizziness” and “vertigo” as a primary or a secondary consulting reason at a primary care setting. We did not limit our search to the date of publications, patient age or study quality. All types of studies except qualitative studies, case reports and narrative reviews could be included.

The findings of the studies had to include at least one of the following estimates: prevalence or incidence of dizziness and/or vertigo, information about underlying diagnoses and/or prognoses. There was no restriction regarding to the kinds of data assessment, outcome measurement or the study quality. We excluded studies that evaluated other settings than primary care (e.g. hospital care, emergency centres, secondary care) and studies, in which patients were selected before recruitment, for example because of an increased probability for a specific underlying diagnosis.

In January 2015, we did a computer-based search of the PUBMED and EMBASE databases. The authors also screened the reference list of all relevant studies (snow ball search). Studies published in English, German, Dutch, Italian, Russian, Swedish, French and Spanish were included.

We used the following search syntax for electronic searches:

We defined two main search concepts (“dizziness” and “primary care”) and combined them by “AND”. To operationalize “dizziness”, we used the terms “dizziness”, “vertigo”, “giddiness” and “light headedness” in different ways of notations (in title or abstract) OR the MESH terms “dizziness” and “vertigo”. To operationalize “primary care”, we used four different strategies and combined them by OR: (1) the terms “primary care”, “general practice” or “general practitioner” in different ways of notations in title or abstract, (2) the MESH terms “family practice”, “physicians, family”, “primary Health care” and “rural health”, (3) paper was published in a journal, which typically covers primary care research OR (4) the term “primary care” in different ways of notations appeared in the affiliation of at least the main author. The whole syntax can be found in Additional file 1.

All identified studies went through a two-step selection process. First, we screened titles and abstracts. Studies meeting all three selection criteria “original research article”, “inclusion of patients because of dizziness and/or vertigo” and “primary care setting” were applied as potential appropriate. In the next step, we analysed the full texts of the selected articles regarding to inclusion- and exclusion criteria. Reasons for exclusion were documented.

Two independent review authors (SS/PG or SS/LS for title/abstracts; SS/DB or SS/MB for full texts) did the whole selection process. Disagreements were resolved by discussion between the respective two review authors (SS, PG, LS, DB, and MB). A third review author (either JH or SB) was consulted if disagreement persisted.

We extracted bibliographic data (author, publication year, title and journal), country, setting, study design, inclusion- and exclusion criteria, kind of recruitment, study population (age, gender distribution), study duration. To answer the first research question (prevalence/incidence), we registered the number of dizziness cases and the number and type of the population from which the cases descended from (e.g. number of all practice consultations or all registered patients of a practice). Furthermore, we extracted all diagnostic categories and their absolute and relative frequencies (second research question “aetiology”). Finally, we documented every kind of prognostic outcome (third research question). In addition, we differentiated in our sub-analysis, where data allowed doing so, patients with dizziness from patients suffering from vertigo.

To our knowledge, no tools exist yet to assess the risk of bias in symptom evaluating studies. Therefore, we developed a standardized tool, based on the sparse methodological literature [7, 8] and own previous experience in the area [9, 10]. Two reviewers (SS and KH) independently answered signal questions and assessed the risk of bias separately for four key domains: selection of patients and GPs, data collection and patient flow, determination of the underlying aetiology, and determination of the prognosis. The characteristic values were documented with “yes”, “no” or “unclear”. In addition, reviewers rated their concern that the selection of patients and GPs might had introduced clinical heterogeneity. A detailed description of the tool can be found in Additional file 2.

Studies can use different measures to report the prevalence or incidence of the symptom in a particular setting, e.g. patients presenting with the symptom in relation to all patients presenting in a specified time period; number of consultations due to the particular symptom in relation to all consultations in a specified time period etc. Since the definition of the nominator and denominator of the frequency measure has a major influence on the results, we grouped studies according to the frequency measure and plotted the results using forest plots. To answer question 2 (aetiologies and respective frequencies), we aimed to estimate how often dizziness and/or vertigo were caused by a particular aetiology. For each study presenting data for a particular condition, we calculated the respective proportion and the 95% confidence interval using the Wilson procedure with a correction for continuity. [11] We grouped all eligible studies by underlying aetiologies and plotted the results using forest plots.

We used different measures to quantify the variability across studies: I² quantifies the percentage of variation that is not due to chance [12]. Tau² is an estimate of the between-study variance in a random-effects meta-analysis [12]. Note that in our case the term ‘effect’ refers to a proportion, e.g. proportion of patients with a particular condition. The interpretation of tau² is not intuitive. However, it allows the calculation of prediction interval. Prediction intervals can be interpreted as a range within the “true” proportion of a future study that is similar to those included in the analysis will lie with a probability of 95% [13]. Besides the number of studies, the heterogeneity across studies determines the width of the interval.

Since we expected substantial variation across studies due to different sources (methodological heterogeneity as well as clinical heterogeneity caused by different definitions of the symptom, of underlying conditions, differences in the diagnostic and prognostic work-up, case-mix, health care system and time period), we aimed to provide only qualitative summaries instead of pooled estimates.As there was only a limited number of studies that contributed data to our third research question (prognosis), we analysed these data only descriptively.

Data analysis was done by the statistical program R 3.1.2 (R Foundation for statistical analysis, Vienna, Austria).

Our initial search identified 1598 references in EMBASE and 903 in PUBMED. The snowball search identified eleven references. After extraction of 332 duplicates, 2501 unique references remained. The title and abstract screening of these references detected 120 studies as potential appropriate. Finally, the full text analysis of these trials produced 31 studies, which met the inclusion criteria. Further details are given in a flow chart (Fig. 1).

Data accrue mainly from Europe (21 studies) and USA (five studies). Time of publications varied between 1972 and 2014. Twenty three studies took place in general practice, five in primary health care centres and the remaining three in hospital based primary care centres. Studies varied widely concerning the number on included patients with dizziness ranging from 30 to 10,871. The percentage of women included was mentioned in 16 studies ranging from 58 to 80%. Nineteen of the included studies recruited patients prospectively during the consultations; the other studies were based on chart review and routine data. Further details of the included studies are shown in Table 1.

Quality assessment happened for four distinctive domains. For domain A (selection of patients and GPs), 14 studies had a low, three an unclear and 12 a high risk of bias. For domain B (data collection and patient flow), 19 studies had a low, one an unclear and ten a high risk of bias. For domain C (assessment of aetiology), two studies had a low, two an unclear and ten a high risk of bias. For domain D (assessment of prognosis), five studies had a low, two an unclear and one a high risk of bias. Table 2 summarizes the quality assessment of all included studies.

Twentry two studies commented on prevalence data, nine of them had a low risk of bias. Depending on numerator and denominator, results for prevalence can be grouped in four categories. Four studies describe the number of patients who have contacted a physician at least once because of dizziness/vertigo in relation to all listed patients. Prevalence ranges from 0,8% [14] to 7,9% (DNSGP-2) [15]. Seven studies describe the number of patients with dizziness in relation to all patients seen in consultation. Prevalence ranges from 1,2% [16] to 8,1% [17]. Five studies describe the number of consultations due to the symptom dizziness in relation to all consultations (e.g. if the same patient has more than one consultation he is here counted more than one time). Prevalence ranges from 1,0% (BEACH) [18] to 15,5% [19]. In four of the five studies results are below 2%, only the study done by Wun et al. represents with 15,5% an outlier. 50% of the patients seen in this study suffered from hypertension. The authors assume that patients utilized the symptom “dizziness” as a kind of entrance card to get their blood pressure monitored. Finally, four studies describe the number of consultations due to the symptom dizziness in relation to all reasons for encounter. Prevalence ranges from 0,7% (BEACH) [18] to 9,9% [19]. Forest plots and measures of heterogeneity are presented in Additional file 3.

We identified 14 studies assessing data on the aetiology of dizziness and/or vertigo. The most common categories for dizziness were cardiovascular (3,8-56,8%) and otologic peripheral (5,4-42,1%) problems including Benign Positional Paroxysmal Vertigo (BPPV) and vestibular neuritis. Frequent reasons for vertigo were as well otologic peripheral causes, BPPV and vestibular neuritis, whereas cardiovascular disease did not play any role. It is noteworthy that 13 out of 14 studies used a category like “no specific diagnosis possible” and that up to 80% of cases were assigned to that category. Table 3 summarizes the results for several differential diagnoses separated for the symptoms dizziness and vertigo. Forest plots and measures of heterogeneity are presented in Additional file 4.

We could identify only two studies with a low risk of bias. Both looked at older patients with the symptom dizziness and identified cardiovascular disease as main aetiology followed by peripheral vestibular disease (see Table 4).

Prognostic parameters were assessed in eight studies (two of them with a low risk of bias) using different end-points. Three studies all conducted by Yardley et al. [20‐22] used the short form of the vertigo symptom scale (ranging from 0 to 60 points). All studies were intervention studies in the field of vestibular rehabilitation. In none of the studies, the respective control groups showed a significant change of 3 points (5%) at the end of the one-year follow up period.

Five studies [20‐24] looked with the help of different instruments like the dizziness handicap inventory, the vertigo handicap questionnaire or the SF 36 (short version) at changes in quality of life. All studies only showed slight improvement, which was in most instances regarded as clinically non-significant.

Five studies [20, 22, 25‐27] measured subjective patient assessment. Patients reported subjective improvement in al studies ranging from 37% [5] to 77% [26].

This systematic review identified 31 studies of the symptoms dizziness and/or vertigo at the primary care setting. In regard to all research questions results were very heterogeneous. The most common reasons for dizziness were cardiovascular (3,8-56,8%) and otologic peripheral problems (5,4-42,1%). Frequent reasons for vertigo were as well otologic peripheral causes, BPPV and vestibular neuritis, whereas cardiovascular disease did not play any role. In up to 80% of cases no specific diagnosis could be made. Regarding prognosis, most studies only showed slight, often non-significant improvement.

Prevalence of dizziness varied markedly between studies. One reason might be that dizziness as a symptom depends on capturing and judging various sensations [2] and therefore epidemiological research faces difficulties describing and standardizing this symptom [28]. Furthermore, for identification of patients presenting with dizziness some studies counted every dizzy patient prospectively or used databases that provide data on reasons for encounter for every consultation. In contrast, other studies used routine data, with a high risk of overlooking dizzy patients with diagnoses that are not necessarily accompanied by dizziness. Both might explain heterogeneity of prevalence data.

Compared to data from emergency departments, where dizziness accounts for 2,5% [29] and 3,3% [30] of consultations, our findings provide similar figures.

At primary care level GPs do not necessarily need to know the exact diagnoses. This is in accordance with studies based on routine data, where in 46,0 to 80,2% of the cases an explicit diagnosis is missing. For a symptom, that according to Sloane et al. [2] resolves for most patients spontaneously, this seems to be a reasonable approach as long as avoidable life-threading conditions can be reliably identified. On the other hand, GPs describe dizziness as a confusing and difficult problem to deal with [4]. Geser et al. [31] report that in most of the patients referred to a specialized dizziness clinic neuro-otological disorders were underdiagnosed. However, with a diagnostic approach, substantially based on medical history and clinical examination, GPs could be successfully trained to provide a better work up. Our findings support this suggestion: The proportion of unclear diagnoses decreases to 0,0-22,0%, if only prospective studies exclusively on the symptom dizziness were considered, where participating GPs presumably had to deal in one way or another with diagnostic strategies to provide a work up of their patients.

Two reviews, including studies from primary care and other settings find 4 to 48% peripheral vestibular diseases [2] and a quality-adjusted mean for peripheral vestibular diseases of 44% [32]. Studies from specialized dizziness clinics show 40,9% [33] and 46,8% [34] peripheral vestibular disease. These findings are higher than ours (5,4-42,1% peripheral vestibular diseases).

Regarding cardiovascular diseases, we found a higher proportion in older patients, especially in studies with a low risk of bias. For patients of 60 or 65 years and older the two studies with a low risk of bias showed 48,0% [35] and 56,8% ([36] cardiovascular problems. In comparison, the results of studies with a high risk of bias that included only older patients range from 10,0% [4] to 13,0% [24].

It is interesting to note that in one study conducted in the emergency department, 21,1% of dizzy patients suffered from a cardiovascular disease [30, 37], whereas in dizziness clinics cardiovascular diseases do not rank among the mentioned frequent aetiologies [31, 33, 34]. Dizziness due to cardiovascular problems might go along with other symptoms like angina pectoris that rather warrant a visit to the emergency department.

Knowledge of prognosis is helpful in the context of further decision making in primary care. If dizziness in this setting has a good prognosis and will disappear in most patients from alone then it would not be necessary to always make an exact diagnosis provided red flags are taken into account. Symptomatic therapy would suffice.

While in the included studies there could be shown improvement based on subjective patient assessment, this could not be confirmed when applying instruments that measure quality of life.

In comparison, at a specialized dizziness clinic the mean score of the Dizziness Handicap Inventory (DHI) improved significantly after two year follow-up [38]. Possible explanations for this difference could be that treatment interventions in specialized clinics are more effective. Another explanation would be the shorter follow up in the primary care studies.

Overall, there is a scarcity of studies, which investigate prognosis of dizziness in primary care: Two of the four studies using the DHI include only patients of a minimum age of 65 years. The other two are intervention studies. Yardley et al. reported that about 90% of the invited patients did not participate in the study assuming that the majority of these patients were no longer dizzy [20]. This selection effect could be one explanation for the rather bad prognosis in primary care.

The quality of the results of a systematic review depends on the quality of the included studies. Although we found 9 studies of high quality (low risk of bias for the domains A and B), which provide data on prevalence, we decided, that a meta-analysis was not reasonable because of the high clinical and methodical heterogeneity between these studies. For example, studies differ in outcomes (numerator and denominator), duration of dizziness, patient ages, proportion of women, number and type of health care institutions (practices, health care centre). Furthermore, study results depend on cultural variances, regarding, inter alia, health care systems and definition of dizziness in different countries.

Regarding aetiology we found only two studies with a low overall risk of bias (low risk of bias for the domains A, B and C). The majority of studies considers just the GPs diagnosis without follow up or standardised diagnostic approach. Moreover they do not define aetiological categories and therefore the forming of categories was difficult and at risk of overlapping. For example, some studies provided no information, under which category cerebrovascular diseases were grouped (neurological or cardiogenic).

Dizziness may have a multifactorial origin and therefore some studies allow more than one diagnosis per patient, resulting in a sum of all diagnoses over 100% and accordingly higher percentages in all aetiological categories. Other studies have a high proportion of unclear diagnosis, leading to lower percentages in other categories.

Due to the small number of trustworthy studies we could not assess, how duration of dizziness, patients age or gender influenced the proportion of underlying etiologies.

There are limitations to our study, influenced by three factors, which could lead to a bias in interpretation:

First, factors which influence the internal validity of the included studies like incomplete recruitment or imprecise inclusion criteria. We controlled this type of bias with rigorous quality assessment and defined clear inclusion criteria for all included studies.

Second, factors which might influence the external validity of the included studies like setting or recruitment characteristics that influence transferability to the local health care system.

Third, factors which influence the internal validity of this systematic review like mistakes in the screening process or full text analysis. Both processes were performed by two independent reviewers, which minimized this risk.

The majority of the included studies have been conducted in health care systems with a gatekeeper, some studies are from countries with direct access to specialists care. This lack of a filter might have an impact on the aetiology of the symptoms and adds an additional source of heterogeneity. These differences cannot only be seen between countries but also within one country. Germany, for example, has not an official gate keeping system but some counties have implemented a voluntary gate keeping system. Additionally, in rural areas patients tend always to contact their GP first while in urban areas specialist are contacted directly in comparatively higher frequency.

Due to the missing quality standard for symptom evaluating studies we developed a more comprehensive catalogue of criteria that, when applied to the included studies, showed a broad spectrum from low to high risk of bias. As this also may introduce variance across studies, we did not pool data.

Although we have decided on structured steps concerning our search strategy, data extraction and analysis, we have not published a review protocol.