Effectiveness of training on vertigo/dizziness coding in primary care: a community-based cluster-randomized trial Authors: Jenniffer Perez-Patiño, Oriol Cunillera-Puértolas, Albert Sanllorente-Melenchón, Yolanda Rando-Matos, Iván Villar-Balboa, Ricard Carrillo-Muñoz, José Luis Ballvé-Moreno, Eva Peguero-Rodríguez Published: 2026 ABSTRACT BackgroundMost patients with vertigo receive unspecific diagnoses, delaying treatment and causing unnecessary tests. Training primary care (PC) physicians is essential to improve accuracy. This study evaluates the effectiveness of an online training program with a face-to-face session on new specific diagnoses in PC.MethodsA multicenter, open-label, cluster-randomized controlled community trial (control group [CG]/ intervention group [IG]) in twenty PC centers in southern Barcelona province (10 per group) was conducted to evaluate the effectiveness of the intervention under real-world primary care conditions. IG professionals received 14 h of online training and one face-to-face session on vertigo management. The primary outcome was the description of new specific vertigo diagnoses. Secondary outcomes included the unspecific, and any vertigo/dizziness diagnoses, the description of individual specific diagnoses (benign paroxysmal positional vertigo [BPPV], Meniere’s disease, vestibular neuritis), unspecific diagnoses (dizziness and other peripheral vertigos) and sociodemographic characteristics obtained from the SIDIAP (Information System for the Development of Primary Care Research) database for 545,938 patients in 2019 and 513,638 in 2022.ResultsIn 2019, the mean age was 51.0 years; 51.3% were women, and 2.68% of patients received a new vertigo-related diagnosis (2.44% nonspecific, 0.28% specific). In 2022, after the intervention, this increased to 2.80% (2.59% nonspecific, 0.28% specific). The IG showed significantly greater use of specific diagnoses (0.30% vs. 0.25%, p = 0.003), particularly BPPV (0.27% vs. 0.23%, p = 0.005). Regression models confirmed higher odds of diagnosis in the IG for any vertigo code (OR = 1.061), specific diagnoses (OR = 2.280), especially BPPV (OR = 2.283).ConclusionA structured training intervention significantly improved the diagnostic coding of vertigo in primary care, fostering more accurate and specific diagnoses.Trial registrationNCT04929444. First Submitted: 07/06/2021.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12909-026-09645-2. Background Around 5% of primary care (PC) consultations are for dizziness, and 20–30% of people will seek medical attention for this reason at some point in their lives [1–3]. Vertigo is its most frequent cause and significantly impacts quality of life and healthcare systems [4]. Despite its prevalence, around 75% of vertigo consultations receive unspecific diagnoses such as "dizziness and giddiness" (ICD-10 R42) [5]. In PC, this figure may reach 80%, though it can drop to 22% with appropriate history and examination [6]. The significant clinical difficulty in diagnosing patients with vertigo or dizziness, due to the wide range of potential causes — many of which are non-otolaryngological, such as central vascular events, neurological, cardiovascular, and psychiatric disorders — may contribute to the use of unspecific diagnoses [3, 7–9]. In the literature, specific diagnoses refer to conditions with defined diagnostic criteria and pathophysiology (e.g., BPPV, Ménière disease, vestibular neuritis), whereas unspecific diagnoses correspond to symptom-based coding when no clear etiology can be established [10, 11]. In both PC and emergency care, diagnostic approaches are often suboptimal, leading to imprecise coding [2, 12, 13]. The German Dizziness and Balance Disorders Center (DSGZ) attributed this to fragmented knowledge, inconsistent guidelines, poor collaboration, and difficulty integrating insights from other fields [3]. Benign paroxysmal positional vertigo (BPPV) accounts for 39% of vertigo cases and can be diagnosed and treated effectively at bedside, avoiding unnecessary investigations [14]. Likewise, accurate diagnosis is essential for correct treatment, in vestibular migraine, that also has a specific treatment [15]. Structured diagnostic tools, such as integrated clinical algorithms and the systematic application of maneuvers like HINTS, have been shown to improve diagnostic accuracy and reduce the proportion of patients without a definitive diagnosis in emergency departments [16–18]. Also, can help reduce diagnostic delays and unnecessary testing, especially given the clinical challenges and the broad range of non-otolaryngological causes underlying vertigo and dizziness [8, 9].In view of primary care’s territorial nature and new educational tools, evaluating knowledge transfer methods is essential. Blended learning may offer a suitable approach, though no studies have assessed its impact on improving vertigo coding. We hypothesised that the training intervention would be associated with a pre–post change in the probability of receiving a specific vertigo diagnosis at the population level. Separately, we examined pre–post changes in the probability of receiving an unspecific vertigo diagnosis. The primary objective of this study was to assess the effectiveness of a clinical training intervention—delivered online through clinical cases with author support and reinforced by a practical in-person session—in improving the diagnosis on new specific diagnoses of vertigo in PC. The secondary objective was to assess the description of unspecific, and any vertigo/dizziness diagnoses and of specific individual diagnoses (BPPV, Ménière's disease and vestibular neuritis) and unspecific diagnoses (dizziness and other peripheral vertigos). Also, sociodemographic characteristics were measured. Methods Design and study settings This multicenter, open-label, community-based, cluster-randomized controlled trial was conducted in 20 PC teams (PCTs) from the Metropolitan South Area of the Catalan Institute of Health (ICS) in Barcelona. Of the 56 PCTs in the area, those that had received vertigo training within the previous five years were excluded prior to randomization. A two-stage randomization process was then performed: firstly, 30 of the remaining PCTs were randomly selected; secondly, these were allocated to either the intervention group (IG) or the control group (CG), with 10 PCTs in each group and an additional five per group to replace any centers declining participation. No stratification was used. The director of the PCT was invited to participate via email. While there were no explicit refusals, there was no response from ten centres in total, five from the intervention group and five from the control group. As only information from participating centres was available, no comparison could be made with non-participating centres. The randomization sequence was generated using the RAND function in MySQL and executed by an independent data manager from SIDIAP (Information System for the Development of Research in PC). Centers in which fewer than 60% of staff consented to participate were excluded. The first ten eligible PCTs in each group formed the final study sample. SIDIAP enrolled the clusters and assigned them to the study arms, maintaining access to the randomization sequence but remaining uninvolved in subsequent data analysis or interpretation. Due to the nature of the educational intervention, investigators were aware of group allocation in order to deliver the intervention. However, outcome data were extracted from collected electronic health records, and data analysis was performed on anonymized datasets in which group allocation was masked to the analysts until the primary analyses were completed. The trial was designed as a superiority study to determine whether the intervention was more effective than usual care. The study population included all patients aged over 14 years seen by General Practitioners (GPs) in the selected PCTs. Data were collected for a pre-intervention period (January to December 2019) and a post-intervention period (January to December 2022). According to the trial registry, the pre-intervention period was initially planned for 2021. However, due to the COVID-19 pandemic and the resulting disruption of routine Primary Care activity, including reduced face-to-face consultations, time constraints, and incomplete diagnostic coding in electronic health records, the study timeline was revised. The intervention was delivered between October and December of 2021, and the pre-intervention assessment was therefore conducted using data from 2019, the last full year preceding the pandemic, when clinical activity and diagnostic recording reflected routine practice. Using 2019 as the baseline allowed for a more reliable and comparable assessment of diagnostic patterns, minimising bias related to pandemic-related changes in healthcare utilisation and documentation. This deviation from the registered protocol was based on theoretical and methodological considerations related to the exceptional impact of the pandemic and was decided prior to any analysis of the outcome data; no analyses were conducted using the originally planned pre-intervention period. The modification is reported transparently as a methodological amendment intended to preserve the original study rationale. The trial was registered with the number NCT04929444 in ClinicalTrials.gov on July 29th 2022 (https://pubmed.ncbi.nlm.nih.gov/35906606/). Further methodological details are available in the study protocol previously publicated [19]. Informed consent was obtained from the physicians of the 20 PCTs study. Patients weren't involved in the design, conduct and reporting of the trial. A pilot study was not considered, as this is a study conducted in routine medical practice aiming to assess whether the intervention produces changes in diagnostic coding. Intervention Clusters corresponded to PCTs, and the educational intervention was offered to all family physicians working within the selected teams. Participation in the educational intervention was voluntary. Family physicians who attended the initial center presentations and provided contact details were invited to participate in the online course. Physicians in the IG received training at the start of the study. This consisted of a 14-h online course on vertigo management, with an estimated time commitment of six weeks from the moment the professional registered and logged in, combining theoretical content and clinical cases, including specially produced videos and immediate feedback on answers. The online course modules opened sequentially upon completion of the previous one; there was no set time limit for each module. There were nine theoretical modules (General Introduction to the Management of Patients with Dizziness and Vertigo, Benign Paroxysmal Positional Vertigo, Vestibular Migraine, Ménière's Disease, Vestibular Neuronitis, Psychogenic Vertigo: Persistent Postural-Perceptual Dizziness, Central Instability, Orthostatic Hypotension). The final grade was pass or fail, based on the requirement that the student read the material and achieve a minimum score of 80% on a multiple-choice exam. An online forum allowed for questions during the course. This part took place from October to December 2021. Following this, a 90-min in-person session was held where participants practiced and standardized diagnostic and therapeutic maneuvers. No formal assessment was performed; instead, participants’ performance of the maneuvers was observed by the investigator, who provided feedback and correction when necessary. This part was carried out in December 2021. The intervention was delivered by trained personnel specialists in PC with experience in vertigo management, and its format (online modules plus an in-person session) was chosen for its feasibility and alignment with local professional development practices [19]. The CG professionals continued with their usual clinical practice throughout the study and were offered the training upon its conclusion. Outcomes and data sources Data were sourced from the SIDIAP which contains pseudonymized health records from ICS primary care, covering around 75% of the Catalan population [20]. Data from 2019 and 2022 were collected for all patients assigned to the participating PCTs. The primary outcome was having any registration of specific new diagnoses (separately; baseline data and during the year following the intervention, from January to December 2019 and 2022, respectively) according to ICD-10: BPPV (H81.1), Meniere’s disease (H81), and vestibular neuritis (H81.2) were categorized as specific (Supplementary Table 1). Additionally, secondary outcomes included the description of unspecific, and any vertigo/dizziness diagnoses and of specific individual diagnoses (BPPV, Ménière’s disease, vestibular neuritis) and the description of individual unspecific diagnoses (dizziness and other peripheral vertigos). ICD-10 codes: (R42) and other peripheral vertigos (H81.3) were classified as unspecific (Supplementary Table 1). Also sex, age, data on patients’ socioeconomic deprivation (MedeA index, categorized into rural and five urban quintiles) and nationality were collected. Recurrent episodes of a same diagnose type in the same patient during the same period were considered once, as having a registration of such diagnosis during the period. These diagnostic categories were prespecified in the study protocol and analyzed separately to capture the full spectrum of vertigo-related diagnostic coding in PC. Sample size The original sample size calculation, described in the published protocol, was based on detecting an improvement in diagnostic accuracy among patients with vertigo. For the present primary outcome, which focused on the incidence of newly recorded vertigo and dizziness diagnoses, sample size considerations were informed by published annual incidence estimates of vestibular vertigo (1.4%) [21]. Based on a cluster randomized design with 10 primary care teams per group, an intracluster correlation coefficient (ICC) of 0.01 [22], and a cluster size exceeding 200 individuals, the minimum detectable absolute difference from a baseline cumulative incidence of 1.4% was estimated to be 1.04%, assuming a two-sided alpha of 0.05 and 80% power. Statistical analysis Data was analyzed according to the Consort Cluster guideline (http://www.equator-network.org/reporting-guidelines/consort-cluster/), and an intention-to-treat for comparisons between the groups was used. Baseline characteristics were described using frequencies and percentages; Chi-square tests assessed group differences regarding study variables. All analyses were patient-based, even those referencing physicians or centers. No missing data was handled, as outcomes were the presence of diagnostic codes, any "missing data" being considered as the absence of a diagnostic code. Follow-up data included all active patients as of 1 January 2022. Mixed logistic regression models evaluated the impact of the intervention on diagnostic coding, adjusting for year, group allocation, and their interaction, and accounting for intra-individual correlations; models considering both individual correlation and PCT were adjusted as sensitivity analysis, as well as models considering no correlation. A significance level of α = 0.05 was used when interpreting p-values. Confidence intervals are presented at the 95% level and were not adjusted to match this corrected significance threshold, as they are intended to provide a classical measure of effect size and precision consistent with the literature [22]. Analyses were conducted using R (v4.1.2 R Foundation for Statistical Computing, Vienna, Austria). Results CONSORT flow diagram is shown in Fig. 1. Of 1,334,381 patients, 30 PCTs were randomized. In the 20 that participated, 97,908 born after 31 December 1999 and 1,857 with missing records were excluded, yielding 595,939 patients for analysis. Workforce size across the participating teams varied over time due to routine staffing changes. Administrative workforce data obtained from Human Resources indicated approximately 314 family physicians in 2019 and 354 in 2022 across the 20 participating PCTs. No formal cluster (PCTs) replacement occurred during the study period. These data are provided in Supplementary Table 2.Fig. 1CONSORT 2025 flow diagram CONSORT 2025 flow diagram The intervention and comparator were delivered as planned by trained personnel. Regarding participation and adherence, 117 of 139 eligible family physicians enrolled in the online course; 103 completed the baseline questionnaire and 73 completed the full 14-h programme. Attendance at the face-to-face session was high (110 participants, 79%). Engagement with the online components was monitored through forum activity (14 questions submitted). Although detailed metrics such as time spent per module were not recorded, module completion was mandatory, and no relevant deviations from the planned intervention were identified. During the six-week online course, one of the course authors made seven follow-up announcements. In the 'I have questions' section of the forum, there were 14 questions, most of which were related to the technical aspects of the study materials, the assessment and the diploma. Seven of the questions were about the theory. Both groups continued to have the usual tools available in clinical practice. In 2019, the participating PCTs had a total assigned population of 545,938 individuals—284,150 assigned to the IG and 261,788 to the CG. The mean age was 51.03 years; 51.29% were women, 84.45% held Spanish nationality, and there were lower proportions of the most and least socioeconomically deprived quintiles (10.10% and 10.06%, respectively) (Table 1). Among healthcare professionals, 89.92% were family medicine specialists; 62.17% were women, 29.41% were over 55 years old, and only 0.66% were under 30 (Table 2). In 2022, after the intervention period, the total population was 513,638 (268,082 in the IG and 245,556 in the CG), with a mean age of 50.18 years, a 51.47% were women and the 85.96% of population were Spanish (Table 1). The 95.58% of professionals were GP's, the 66.09% were women, and the 31.26% were over 55 years old whereas only 0.7% were under 30 (Table 2). In 2019, 2.68% of the population received a new vertigo-related diagnosis: 2.44% were unspecific and 0.28% specific (Table 3). The most common diagnostic code was the unspecific “dizziness” (2.05%), followed by the specific “BPPV” (0.26%). By 2022, the overall proportion of new vertigo-related diagnoses rose slightly to 2.80% (2.59% unspecific, 0.28% specific). “Dizziness” remained the most frequent code (2.00%), followed by unspecified vestibular disorder (0.30%) and “BPPV” (0.26%). In 2019, although group differences in coding were statistically significant for BPPV and overall specific diagnoses, they were not clinically relevant (e.g. BPPV: 0.25% IG vs 0.28% CG, p = 0.015; any specific: 0.27% IG vs 0.30% CG, p = 0.034). In 2022, statistically significant increases were observed in the IG for specific diagnoses overall (0.30% vs 0.25%, p = 0.003), BPPV (0.27% vs 0.23%, p = 0.005), and also for the unspecific code “dizziness” (2.05% vs 1.94%, p = 0.003).Table 1Baseline patient characteristics of the active sample in 2019 and 2022 by exposure groupGlobal (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p valuePatients 2019Patients 2022Sex< 0.001 Women280,028 (51.29%)144,872 (50.98%)135,156 (51.63%)< 0.001 Men265,910 (48.71%)139,278 (49.02%)126,632 (48.37%)264,348 (51.47%)137,148 (51.16%)127,200 (51.80%)Age at January first (mean, SD)51.03 (17.80)50.89 (17.74)51.18 (17.87)< 0.001249,290 (48.53%)130,934 (48.84%)118,356 (48.20%)Age groups< 0.00150.18 (17.13)50.05 (17.08)50.32 (17.19)< 0.001 < = 50287,320 (52.63%)150,805 (53.07%)136,515 (52.15%)< 0.001 (50, 75]200,879 (36.80%)103,988 (36.60%)96,891 (37.01%)277,310 (53.99%)145,922 (54.43%)131,388 (53.51%) > 7557,739 (10.58%)29,357 (10.33%)28,382 (10.84%)191,905 (37.36%)99,503 (37.12%)92,402 (37.63%)Nationality< 0.00144,423 (8.65%)22,657 (8.45%)21,766 (8.86%) Unmeasured37,981 (6.96%)23,760 (8.36%)14,221 (5.43%)< 0.001 Spain468,360 (85.79%)242,021 (85.17%)226,339 (86.46%)36,039 (7.02%)22,653 (8.45%)13,386 (5.45%) Eastern Europe5,104 (0.93%)2,749 (0.97%)2,355 (0.90%)441,544 (85.96%)228,682 (85.30%)212,862 (86.69%) Europe, other6,329 (1.16%)2,097 (0.74%)4,232 (1.62%)4,644 (0.90%)2,521 (0.94%)2,123 (0.86%) America11,014 (2.02%)5,004 (1.76%)6,010 (2.30%)5,662 (1.10%)1,888 (0.70%)3,774 (1.54%) Northern Africa10,984 (2.01%)5,607 (1.97%)5,377 (2.05%)9,943 (1.94%)4,512 (1.68%)5,431 (2.21%) Africa, other1,478 (0.27%)569 (0.20%)909 (0.35%)10,146 (1.98%)5,152 (1.92%)4,994 (2.03%) Southern Asia1,862 (0.34%)878 (0.31%)984 (0.38%)1,312 (0.26%)496 (0.19%)816 (0.33%) Asia, other2,801 (0.51%)1,456 (0.51%)1,345 (0.51%)1,702 (0.33%)806 (0.30%)896 (0.36%) Australia, NZ and Polynesia25 (0.00%)9 (0.00%)16 (0.01%)2,625 (0.51%)1,364 (0.51%)1,261 (0.51%)Socioeconomic deprivation index< 0.00121 (0.00%)8 (0.00%)13 (0.01%) Unknown Urban83,519 (15.30%)45,315 (15.95%)38,204 (14.59%)< 0.001 Rural62,681 (11.48%)54,119 (19.05%)8,562 (3.27%)67,263 (13.10%)36,973 (13.79%)30,290 (12.34%) Least deprived55,147 (10.10%)17,063 (6.00%)38,084 (14.55%)60,500 (11.78%)52,163 (19.46%)8,337 (3.40%) Q285,244 (15.61%)50,693 (17.84%)34,551 (13.20%)53,350 (10.39%)16,665 (6.22%)36,685 (14.94%) Q3116,183 (21.28%)61,019 (21.47%)55,164 (21.07%)82,432 (16.05%)49,090 (18.31%)33,342 (13.58%) Q488,263 (16.17%)39,323 (13.84%)48,940 (18.69%)111,962 (21.80%)59,060 (22.03%)52,902 (21.54%) Most deprived54,901 (10.06%)16,618 (5.85%)38,283 (14.62%)85,213 (16.59%)38,028 (14.19%)47,185 (19.22%)Table 2Baseline healthcare professional characteristics of the active sample in 2019 and 2022 by exposure groupGlobal (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p valueHealthcare professional 2019Healthcare professional 2022Sex< 0.001< 0.001 Female339,434 (62.17%)178,422 (62.79%)161,012 (61.50%)339,462 (66.09%)178,436 (66.56%)161,026 (65.58%) Male148,845 (27.26%)73,127 (25.74%)75,718 (28.92%)148,856 (28.98%)73,131 (27.28%)75,725 (30.84%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data12,023 (2.20%)9,885 (3.48%)2,138 (0.82%)12,026 (2.34%)9,888 (3.69%)2,138 (0.87%)Age ranges< 0.001< 0.001 < 30 years3,620 (0.66%)1,132 (0.40%)2,488 (0.95%)3,620 (0.70%)1,132 (0.42%)2,488 (1.01%) [30, 45)173,066 (31.70%)97,630 (34.36%)75,436 (28.82%)173,076 (33.70%)97,636 (36.42%)75,440 (30.72%) [45, 55)151,041 (27.67%)75,848 (26.69%)75,193 (28.72%)151,060 (29.41%)75,857 (28.30%)75,203 (30.63%) > = 55160,552 (29.41%)76,939 (27.08%)83,613 (31.94%)160,562 (31.26%)76,942 (28.70%)83,620 (34.05%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data12,023 (2.20%)9,885 (3.48%)2,138 (0.82%)12,026 (2.34%)9,888 (3.69%)2,138 (0.87%)Specialist< 0.001< 0.001 Family medicine490,914 (89.92%)253,190 (89.10%)237,724 (90.81%)490,954 (95.58%)253,209 (94.45%)237,745 (96.82%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data9,388 (1.72%)8,244 (2.90%)1,144 (0.44%)9,390 (1.83%)8,246 (3.08%)1,144 (0.47%)Table 3New diagnoses by categories, according to exposure group in 2019 and 2022Global (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p value2019 New diagnoses2022 New diagnosesBPPV1,434 (0.26%)700 (0.25%)734 (0.28%)0.0151,310 (0.26%)735 (0.27%)575 (0.23%)0.005Meniere 's Syndrome82 (0.02%)50 (0.02%)32 (0.01%)0.13281 (0.02%)46 (0.02%)35 (0.01%)0.474Vestibular neuritis14 (0.00%)5 (0.00%)9 (0.00%)0.33924 (0.00%)15 (0.01%)9 (0.00%)0.420Central vertigo8 (0.00%)5 (0.00%)3 (0.00%)0.8128 (0.00%)3 (0.00%)5 (0.00%)0.632Other specifics7 (0.00%)3 (0.00%)4 (0.00%)0.91410 (0.00%)6 (0.00%)4 (0.00%)0.859Any specific diagnose1,533 (0.28%)756 (0.27%)777 (0.30%)0.0341,423 (0.28%)800 (0.30%)623 (0.25%)0.003Unspecific disorder of vestibular function99 (0.02%)71 (0.02%)28 (0.01%)< 0.0011,561 (0.30%)800 (0.30%)761 (0.31%)0.467Dizziness and vertigo11,208 (2.05%)5,914 (2.08%)5,294 (2.02%)0.12710,251 (2.00%)5,499 (2.05%)4,752 (1.94%)0.003Other non-specific2,299 (0.42%)1,162 (0.41%)1,137 (0.43%)0.1542,187 (0.43%)1,076 (0.40%)1,111 (0.45%)0.005Any unspecific diagnose13,332 (2.44%)6,986 (2.46%)6,346 (2.42%)0.41513,291 (2.59%)6,999 (2.61%)6,292 (2.56%)0.286Any diagnose14,644 (2.68%)7,624 (2.68%)7,020 (2.68%)0.97914,368 (2.80%)7,585 (2.83%)6,783 (2.77%)0.153Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecificAbbreviations: BPPV Benign paroxysmal positional vertigo Table 1Baseline patient characteristics of the active sample in 2019 and 2022 by exposure groupGlobal (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p valuePatients 2019Patients 2022Sex< 0.001 Women280,028 (51.29%)144,872 (50.98%)135,156 (51.63%)< 0.001 Men265,910 (48.71%)139,278 (49.02%)126,632 (48.37%)264,348 (51.47%)137,148 (51.16%)127,200 (51.80%)Age at January first (mean, SD)51.03 (17.80)50.89 (17.74)51.18 (17.87)< 0.001249,290 (48.53%)130,934 (48.84%)118,356 (48.20%)Age groups< 0.00150.18 (17.13)50.05 (17.08)50.32 (17.19)< 0.001 < = 50287,320 (52.63%)150,805 (53.07%)136,515 (52.15%)< 0.001 (50, 75]200,879 (36.80%)103,988 (36.60%)96,891 (37.01%)277,310 (53.99%)145,922 (54.43%)131,388 (53.51%) > 7557,739 (10.58%)29,357 (10.33%)28,382 (10.84%)191,905 (37.36%)99,503 (37.12%)92,402 (37.63%)Nationality< 0.00144,423 (8.65%)22,657 (8.45%)21,766 (8.86%) Unmeasured37,981 (6.96%)23,760 (8.36%)14,221 (5.43%)< 0.001 Spain468,360 (85.79%)242,021 (85.17%)226,339 (86.46%)36,039 (7.02%)22,653 (8.45%)13,386 (5.45%) Eastern Europe5,104 (0.93%)2,749 (0.97%)2,355 (0.90%)441,544 (85.96%)228,682 (85.30%)212,862 (86.69%) Europe, other6,329 (1.16%)2,097 (0.74%)4,232 (1.62%)4,644 (0.90%)2,521 (0.94%)2,123 (0.86%) America11,014 (2.02%)5,004 (1.76%)6,010 (2.30%)5,662 (1.10%)1,888 (0.70%)3,774 (1.54%) Northern Africa10,984 (2.01%)5,607 (1.97%)5,377 (2.05%)9,943 (1.94%)4,512 (1.68%)5,431 (2.21%) Africa, other1,478 (0.27%)569 (0.20%)909 (0.35%)10,146 (1.98%)5,152 (1.92%)4,994 (2.03%) Southern Asia1,862 (0.34%)878 (0.31%)984 (0.38%)1,312 (0.26%)496 (0.19%)816 (0.33%) Asia, other2,801 (0.51%)1,456 (0.51%)1,345 (0.51%)1,702 (0.33%)806 (0.30%)896 (0.36%) Australia, NZ and Polynesia25 (0.00%)9 (0.00%)16 (0.01%)2,625 (0.51%)1,364 (0.51%)1,261 (0.51%)Socioeconomic deprivation index< 0.00121 (0.00%)8 (0.00%)13 (0.01%) Unknown Urban83,519 (15.30%)45,315 (15.95%)38,204 (14.59%)< 0.001 Rural62,681 (11.48%)54,119 (19.05%)8,562 (3.27%)67,263 (13.10%)36,973 (13.79%)30,290 (12.34%) Least deprived55,147 (10.10%)17,063 (6.00%)38,084 (14.55%)60,500 (11.78%)52,163 (19.46%)8,337 (3.40%) Q285,244 (15.61%)50,693 (17.84%)34,551 (13.20%)53,350 (10.39%)16,665 (6.22%)36,685 (14.94%) Q3116,183 (21.28%)61,019 (21.47%)55,164 (21.07%)82,432 (16.05%)49,090 (18.31%)33,342 (13.58%) Q488,263 (16.17%)39,323 (13.84%)48,940 (18.69%)111,962 (21.80%)59,060 (22.03%)52,902 (21.54%) Most deprived54,901 (10.06%)16,618 (5.85%)38,283 (14.62%)85,213 (16.59%)38,028 (14.19%)47,185 (19.22%) Baseline patient characteristics of the active sample in 2019 and 2022 by exposure group Table 2Baseline healthcare professional characteristics of the active sample in 2019 and 2022 by exposure groupGlobal (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p valueHealthcare professional 2019Healthcare professional 2022Sex< 0.001< 0.001 Female339,434 (62.17%)178,422 (62.79%)161,012 (61.50%)339,462 (66.09%)178,436 (66.56%)161,026 (65.58%) Male148,845 (27.26%)73,127 (25.74%)75,718 (28.92%)148,856 (28.98%)73,131 (27.28%)75,725 (30.84%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data12,023 (2.20%)9,885 (3.48%)2,138 (0.82%)12,026 (2.34%)9,888 (3.69%)2,138 (0.87%)Age ranges< 0.001< 0.001 < 30 years3,620 (0.66%)1,132 (0.40%)2,488 (0.95%)3,620 (0.70%)1,132 (0.42%)2,488 (1.01%) [30, 45)173,066 (31.70%)97,630 (34.36%)75,436 (28.82%)173,076 (33.70%)97,636 (36.42%)75,440 (30.72%) [45, 55)151,041 (27.67%)75,848 (26.69%)75,193 (28.72%)151,060 (29.41%)75,857 (28.30%)75,203 (30.63%) > = 55160,552 (29.41%)76,939 (27.08%)83,613 (31.94%)160,562 (31.26%)76,942 (28.70%)83,620 (34.05%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data12,023 (2.20%)9,885 (3.48%)2,138 (0.82%)12,026 (2.34%)9,888 (3.69%)2,138 (0.87%)Specialist< 0.001< 0.001 Family medicine490,914 (89.92%)253,190 (89.10%)237,724 (90.81%)490,954 (95.58%)253,209 (94.45%)237,745 (96.82%) Unassigned professional45,636 (8.36%)22,716 (7.99%)22,920 (8.76%)13,294 (2.59%)6,627 (2.47%)6,667 (2.72%) Unavailable data9,388 (1.72%)8,244 (2.90%)1,144 (0.44%)9,390 (1.83%)8,246 (3.08%)1,144 (0.47%) Baseline healthcare professional characteristics of the active sample in 2019 and 2022 by exposure group Table 3New diagnoses by categories, according to exposure group in 2019 and 2022Global (n = 545,938)Intervention (n = 284,150)Control (n = 261,788)p valueGlobal (n = 513,638)Intervention (n = 268,082)Control (n = 245,556)p value2019 New diagnoses2022 New diagnosesBPPV1,434 (0.26%)700 (0.25%)734 (0.28%)0.0151,310 (0.26%)735 (0.27%)575 (0.23%)0.005Meniere 's Syndrome82 (0.02%)50 (0.02%)32 (0.01%)0.13281 (0.02%)46 (0.02%)35 (0.01%)0.474Vestibular neuritis14 (0.00%)5 (0.00%)9 (0.00%)0.33924 (0.00%)15 (0.01%)9 (0.00%)0.420Central vertigo8 (0.00%)5 (0.00%)3 (0.00%)0.8128 (0.00%)3 (0.00%)5 (0.00%)0.632Other specifics7 (0.00%)3 (0.00%)4 (0.00%)0.91410 (0.00%)6 (0.00%)4 (0.00%)0.859Any specific diagnose1,533 (0.28%)756 (0.27%)777 (0.30%)0.0341,423 (0.28%)800 (0.30%)623 (0.25%)0.003Unspecific disorder of vestibular function99 (0.02%)71 (0.02%)28 (0.01%)< 0.0011,561 (0.30%)800 (0.30%)761 (0.31%)0.467Dizziness and vertigo11,208 (2.05%)5,914 (2.08%)5,294 (2.02%)0.12710,251 (2.00%)5,499 (2.05%)4,752 (1.94%)0.003Other non-specific2,299 (0.42%)1,162 (0.41%)1,137 (0.43%)0.1542,187 (0.43%)1,076 (0.40%)1,111 (0.45%)0.005Any unspecific diagnose13,332 (2.44%)6,986 (2.46%)6,346 (2.42%)0.41513,291 (2.59%)6,999 (2.61%)6,292 (2.56%)0.286Any diagnose14,644 (2.68%)7,624 (2.68%)7,020 (2.68%)0.97914,368 (2.80%)7,585 (2.83%)6,783 (2.77%)0.153Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecificAbbreviations: BPPV Benign paroxysmal positional vertigo New diagnoses by categories, according to exposure group in 2019 and 2022 Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecific Abbreviations: BPPV Benign paroxysmal positional vertigo According to the logistic regression models, the intervention led to a population-level increase in the use of diagnostic coding, with a slight rise in the use of any diagnosis (OR = 1.061; p = 0.034) and a substantial increase in the use of specific diagnoses (OR = 2.280; p < 0.001) (Table 4). Specifically, a significant association was observed for increased use of BPPV (OR = 2.283; p < 0.001) and dizziness (OR = 1.109; p < 0.001), along with a decrease in the use of other unspecific diagnoses (OR = 0.839; p = 0.020) and Unspecific disorder of vestibular function (OR = 0.412; p < 0.001). The model for vestibular neuritis proved highly unstable and inconclusive. Tables with correlation by individual and center (PCT), with correlation by center (PCT) and without correlation are shown in Supplementary Tables 3–5. Adjusted sensitivity analyses including sex, age, country, and socioeconomic deprivation index are provided in Supplementary Tables 6–8 to assess the robustness of the findings.Table 4Effect of the intervention on diagnostic coding by category and type, based on logistic regression models with correlationBaseline 2019 in CG2022 change in CGIG2019–2022 Change in IG*BPPV0.000 (0.000, 0.000)< 0.0010.551 (0.550, 0.551)< 0.0010.726 (0.725, 0.726)< 0.0012.283 (2.282, 2.284)< 0.001Meniere's Syndrome0.000 (0.000, 0.000)< 0.0011.166(0.733, 1.754)0.5161.440 (0.949, 2.183)0.0860.836 (0.459, 1.522)0.558Vestibular neuritis0.000 (0.000, 0.000)< 0.0011.011 (0.223, 4.590)0.9890.012 (0.001, 0.165)< 0.001194.511 (18.650, 2028.689)< 0.001Central vertigo0.000 (0.000, 0.000)< 0.0011.777 (0.492, 6.419)0.3801.536 (0.422, 5.591)0.5150.358 (0.061, 2.104)0.256Other specific0.000 (0.000, 0.000)< 0.0011.066(0,318 3,573)0,9170.691 (0.177, 2.703)0.5951.988 (0.345, 11.466)0.442Dizziness and vertigo0.000 (0.000, 0.000)< 0.0010.849 (0.849, 0.849)< 0.0010.991 (0.990, 0,991)< 0.0011.109 (1.048, 1.173)< 0.001Other non-specific0.000 (0.000, 0.000)< 0.0011.044(1.044, 1.044)< 0.0010.994 (0.885, 1,155)0.9360.839(0.723, 0.973)0.020UDVF0.000 (0.000, 0.000)< 0.00129.270 (23.278, 36.804)< 0.0012.337 (1.778, 3.071)< 0.0010.412 (0.312, 0.544)< 0.001Any specific diagnose0.000 (0.000, 0.000)< 0.0010.577 (0.577, 0.577)< 0.0010.699(0.583, 0.839)< 0.0012.280 (1.917, 2.712< 0.001Any unspecific diagnose0.000 (0.000, 0.000)< 0.0011.141 (1.083, 1.202)< 0.0011.018 (1.018, 1.018)< 0.0011.005 (0.937, 1.079)0.879Any diagnose0.000 (0.000, 0.000)< 0.0011.049 (1.049, 1.050)< 0.0010.956 (0.908, 1.007)0.0901.061 (1.004, 1.120)0.034Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecificLogistic regression models were adjusted by year, group allocation, and their interaction; OR < 1 indicates lower odds of code use, OR > 1 higher odds; p ≤ 0.05 indicates a statistically significant effectBaseline 2019 in CG represents the Odds in the control group in 2019, before any interventionChange over time 2019–2022: represents the OR that compares the odds in 2022 with respect to 2019 within the CGBaseline differences in 2019 IG vs CG represents the OR comparing the odds in 2019 between the IG and the CG, before the intervention startedAbbreviations: BPPV Benign paroxysmal positional vertigo, CG Control group, IG Intervention group, UDVF Unspecific disorder of vestibular function*Intervention effect. Change 2019–2022 in IG represents the modifier effect of the intervention; OR of changes between 2019 and 2022 in the IG compared to changes in the CGValues in bold represent statistically significant figures Table 4Effect of the intervention on diagnostic coding by category and type, based on logistic regression models with correlationBaseline 2019 in CG2022 change in CGIG2019–2022 Change in IG*BPPV0.000 (0.000, 0.000)< 0.0010.551 (0.550, 0.551)< 0.0010.726 (0.725, 0.726)< 0.0012.283 (2.282, 2.284)< 0.001Meniere's Syndrome0.000 (0.000, 0.000)< 0.0011.166(0.733, 1.754)0.5161.440 (0.949, 2.183)0.0860.836 (0.459, 1.522)0.558Vestibular neuritis0.000 (0.000, 0.000)< 0.0011.011 (0.223, 4.590)0.9890.012 (0.001, 0.165)< 0.001194.511 (18.650, 2028.689)< 0.001Central vertigo0.000 (0.000, 0.000)< 0.0011.777 (0.492, 6.419)0.3801.536 (0.422, 5.591)0.5150.358 (0.061, 2.104)0.256Other specific0.000 (0.000, 0.000)< 0.0011.066(0,318 3,573)0,9170.691 (0.177, 2.703)0.5951.988 (0.345, 11.466)0.442Dizziness and vertigo0.000 (0.000, 0.000)< 0.0010.849 (0.849, 0.849)< 0.0010.991 (0.990, 0,991)< 0.0011.109 (1.048, 1.173)< 0.001Other non-specific0.000 (0.000, 0.000)< 0.0011.044(1.044, 1.044)< 0.0010.994 (0.885, 1,155)0.9360.839(0.723, 0.973)0.020UDVF0.000 (0.000, 0.000)< 0.00129.270 (23.278, 36.804)< 0.0012.337 (1.778, 3.071)< 0.0010.412 (0.312, 0.544)< 0.001Any specific diagnose0.000 (0.000, 0.000)< 0.0010.577 (0.577, 0.577)< 0.0010.699(0.583, 0.839)< 0.0012.280 (1.917, 2.712< 0.001Any unspecific diagnose0.000 (0.000, 0.000)< 0.0011.141 (1.083, 1.202)< 0.0011.018 (1.018, 1.018)< 0.0011.005 (0.937, 1.079)0.879Any diagnose0.000 (0.000, 0.000)< 0.0011.049 (1.049, 1.050)< 0.0010.956 (0.908, 1.007)0.0901.061 (1.004, 1.120)0.034Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecificLogistic regression models were adjusted by year, group allocation, and their interaction; OR < 1 indicates lower odds of code use, OR > 1 higher odds; p ≤ 0.05 indicates a statistically significant effectBaseline 2019 in CG represents the Odds in the control group in 2019, before any interventionChange over time 2019–2022: represents the OR that compares the odds in 2022 with respect to 2019 within the CGBaseline differences in 2019 IG vs CG represents the OR comparing the odds in 2019 between the IG and the CG, before the intervention startedAbbreviations: BPPV Benign paroxysmal positional vertigo, CG Control group, IG Intervention group, UDVF Unspecific disorder of vestibular function*Intervention effect. Change 2019–2022 in IG represents the modifier effect of the intervention; OR of changes between 2019 and 2022 in the IG compared to changes in the CGValues in bold represent statistically significant figures Effect of the intervention on diagnostic coding by category and type, based on logistic regression models with correlation Clarifications: Other specifics: Labyrinthitis, labyrinthine fistula or labyrinthine dysfunction; Any specific diagnosis: BPPV, Meniere's syndrome, vestibular neuritis, central vertigo, other specific; Other unspecific: other types of peripheral vertigo, other vestibular function disorders, aural vertigo or vertiginous syndromes in diseases classified elsewhere; Any unspecific diagnosis: dizziness and vertigo, unspecified disorder of vestibular function or other unspecific; Any diagnosis: specific + unspecific Logistic regression models were adjusted by year, group allocation, and their interaction; OR < 1 indicates lower odds of code use, OR > 1 higher odds; p ≤ 0.05 indicates a statistically significant effect Baseline 2019 in CG represents the Odds in the control group in 2019, before any intervention Change over time 2019–2022: represents the OR that compares the odds in 2022 with respect to 2019 within the CG Baseline differences in 2019 IG vs CG represents the OR comparing the odds in 2019 between the IG and the CG, before the intervention started Abbreviations: BPPV Benign paroxysmal positional vertigo, CG Control group, IG Intervention group, UDVF Unspecific disorder of vestibular function *Intervention effect. Change 2019–2022 in IG represents the modifier effect of the intervention; OR of changes between 2019 and 2022 in the IG compared to changes in the CG Values in bold represent statistically significant figures Discussion This study explored the impact of a training intervention to GP’s, in improving vertigo care in PC by increasing the use of specific vertigo diagnoses, through a multicenter, open-label, cluster-randomized controlled community trial. Patients in IG were more likely to be diagnosed with accurate and specific diagnosis, compared with those in CG, particularly in BPPV diagnoses. In this study, the increase in specific diagnoses in the IG, particularly BPPV, is notable. As the leading cause of peripheral vertigo, BPPV requires no confirmatory tests and can be diagnosed during consultation using positional manoeuvres [23]. However, posterior canal BPPV and vestibular migraine are frequently misdiagnosed [24]. It is important to note that vestibular migraine is not represented in the ICD-10 coding system used in our electronic health records, and therefore could not be captured, which constitutes a limitation of the administrative data [25]. On the other hand, central vertigo diagnoses (H81.41–H81.43 and H81.49) were included in the “specific” category due to their extremely low frequency in primary care; only eight cases were recorded across the entire study population, as shown in the tables. This grouping reflects both the rarity of central vertigo in the primary care context and the need to maintain analytical parsimony. One explanation for the moderate impact of the intervention in the present study could be that residual changes in care pathways and coding practices may have persisted into 2022, potentially leading to an underestimation of specific diagnoses in the post-intervention period. Studies from Catalan primary care health system indicate that the recovery of face-to-face activity was gradual during 2022–2023, with a late increase in both face-to-face and remote visits in 2022, suggesting that consultation modalities remained dynamic in the post-pandemic period [26]. In addition, previous studies reported reductions of up to 75% in BPPV visits and 88.2% in Ménière’s disease visits in Japan due to fear of contagion [27]. Although the absolute differences at the population level are small (around 0.05 percentage points), the clinical relevance should not be interpreted solely on this basis. In our study, these differences translated into an estimated average of approximately 7 additional specific diagnoses per trained professional, illustrating that the intervention can have a tangible impact at the individual provider level. Such differences imply a high number needed to train to obtain one additional specific diagnosis of BPPV; however, at the individual patient level, a correct etiological diagnosis may have a substantial impact on quality of life by enabling effective treatment and reducing symptom duration [28, 29]. Improved diagnostic accuracy may also reduce unnecessary complementary tests, specialist referrals (ENT or neurology), and prolonged sick leave, all of which have relevant implications for healthcare resource use and costs [30, 31]. Similar educational interventions aimed at improving diagnostic skills in dizziness have generally reported improvements in diagnostic recognition rather than large absolute changes in case numbers, reflecting the relatively low incidence of individual vestibular disorders at the level of individual physicians [32]. Given the high prevalence of dizziness in primary care, even small absolute improvements may have a meaningful cumulative impact at the health system level. Moreover, changes in diagnostic practice are often gradual, and an increase in specific diagnoses may represent an early indicator of a broader shift in clinical approach, as suggested in previous studies where training alone produced modest but relevant effects (e.g., Kerber et al.) [33]. Improving professional training has been proposed as a solution. Professional education has proven useful. A German study emphasized the need to bridge the gap between guidelines and practice via continuing medical education [34]. An Australian study in family medicine residents found that non-specific vertigo diagnoses are frequent, reflecting the diagnostic challenges. The authors recommend reinforcing vertigo training and supervisor involvement in residents' diagnostic reasoning [35]. Similarly, a Spanish study found that 65.6% of patients could be accurately diagnosed in PC by trained physicians using structured history and vestibular examination [36]. A U.S. emergency department study [33] doubled the use of diagnostic and therapeutic maneuvers, yet many BPPV cases remained undiagnosed. Similarly, in our study, BPPV diagnoses more than doubled in the IG, suggesting better diagnostic accuracy, though many cases likely still go unnoticed. Despite these gains, around 90% of vertigo-related cases remained coded with unspecific diagnoses, reflecting the persistent complexity of vestibular disorders in PC. In line with our findings, Murphy et al. [1] likewise found that many patients present overlapping symptoms, highlighting the limitations of current classification systems. These findings reinforce the need for refined diagnostic strategies and decision support tools to improve vertigo management in frontline settings. In summary, improving the clinical diagnosis of vertigo is key to optimizing outcomes and reducing errors, yet significant challenges remain due to the symptom’s etiological heterogeneity and clinical complexity [3, 17, 18]. In our study, despite the emphasis on diagnostic maneuvers—both in the online course and the in-person practical session, where we ensured all students performed them correctly—we were unable to verify their actual use in clinical practice. By comparison, Ledger et al. [35] found that most vertigo diagnoses in Australian general practice were unspecific (68.5%), with specific diagnoses more likely when procedures like repositioning maneuvers were performed or additional advice was sought. However, the number of maneuvers recorded remained much lower than the number of BPPV diagnoses [35]. Previous research supports the importance of enhancing diagnostic skills and therapeutic approaches for vertigo in general practice. Van Vugt et al. [37] found that general practitioners (GPs) who served as trainers were more likely to perform repositioning maneuvers compared to their non-training counterparts, underscoring the relevance of clinical leadership in promoting best practices. Similarly, Hanley and O’Dowd [38] reported that GPs who had received training in these maneuvers could integrate them into routine care with beneficial outcomes. These findings reinforce the need for structured training programs to improve the diagnosis and management of vestibular disorders in PC. Recent research has increasingly focused on improving diagnostic accuracy in patients with vertigo, both through better clinical assessment and supportive tools. Studies such as the VERtigo DIagnosis project [39] and the CAVA trial (Continuous Ambulatory Vestibular Assessment) [40] illustrate ongoing efforts to strengthen diagnostic approaches in this field, although no results are available to date. In contrast, our results reflect the effect of a pragmatic training intervention on diagnostic coding in routine primary care. While training has been shown to improve diagnostic accuracy, evidence suggests that education alone may be insufficient to produce sustained changes in everyday practice. Qualitative studies have identified uncertainty in performing diagnostic maneuvers and interpreting nystagmus as key barriers, with access to high-quality educational resources acting as an important facilitator [41–43]. Consistent with this, a national survey in Switzerland highlighted the potential value of practical courses and digital resources to improve the management of dizziness in primary care [32]. Strengths and limitations Several limitations should be considered when interpreting our findings. Due to the study design, we were unable to determine the proportion of patients who underwent key diagnostic maneuvers, such as the Dix-Hallpike or head impulse test -essential yet often underused in clinical practice—or whether training led to an increase in their use, which limits our ability to link training directly to procedural competence. Time trends and differences between patient groups may also represent potential sources of confounding. The COVID-19 pandemic and subsequent years likely affected diagnostic coding and the use of diagnostic tests for vertigo-related conditions; previous studies reported reductions of up to 75% in BPPV visits and 88.2% in Ménière’s disease visits in Japan due to fear of contagion [27]. To address this, the pre-intervention assessment period was shifted from 2021 to 2019, but this adjustment may introduce time-trend confounding. Moreover, residual changes in care pathways and coding practices may have persisted into 2022, potentially leading to an underestimation of specific diagnoses in the post-intervention period. Despite measures to minimize contamination, such as excluding previously trained centres and restricting access to the intervention, informal professional interactions and physician mobility may have led to some contamination between intervention and control centres, potentially biasing results toward the null. Although absolute differences at the population level were small (~ 0.05%), they may still have meaningful implications for healthcare resource use, as modest improvements in diagnostic accuracy can reduce unnecessary tests, specialist referrals, and prolonged sick leave, not measured in this study. Vestibular migraine could not be assessed due to the absence of a specific ICD-10 code, despite its clinical importance, frequent underdiagnosis, and specific treatment options [24], and central vertigo diagnoses were grouped due to their very low frequency, which may have resulted in some misclassification. In addition, this classification does not capture the full etiological spectrum of vertigo, including cardiovascular, psychiatric and central causes. Given that peripheral vestibular disorders are the most frequent causes of vertigo, the results should be interpreted as reflecting changes in diagnostic coding behaviour for selected peripheral conditions, rather than overall improvements in vertigo diagnostic accuracy. Accordingly, results should be interpreted with caution, placing emphasis not only on statistical significance but also on effect sizes and the consistency of associations observed across outcomes. As the intervention targeted clinicians while outcomes were assessed at the population level, not all patients were attended by trained physicians. In addition, the four-year interval between the 2019 baseline and the 2022 assessment implies that cluster composition may have changed over time due to workforce variability, including physician mobility and turnover. As these changes could not be individually tracked, the results should be interpreted as reflecting a cluster-level effect rather than a direct pre-post comparison of the same clinicians. This pragmatic design may have diluted the intervention effect, however, it reflects real-world implementation conditions and enhances the ecological validity and generalizability of the findings within PC settings. Nonetheless, a key strength of the study is its large sample size, both in terms of patients and healthcare professionals and also carried out in the field of PC. Although professionals were aware of their participation, they retained full autonomy in their clinical decisions, and care was delivered under routine working conditions. Conclusions This study, reflecting real-world PC practice, demonstrates diagnostic improvements in specific vertigo diagnoses following specific training. These gains are particularly evident in selected peripheral vestibular conditions rather than in overall diagnostic accuracy. Furthermore, while the observed differences appear small at the population level, they remain potentially meaningful at the provider level. Likely driven by professionals open to change, these results may inspire others to adopt better approaches. However, many patients still receive unspecific diagnoses. As previous research suggests, training alone improves practice but is insufficient for widespread, lasting change. Achieving consistent, sustainable improvement remains a challenge essential to enhancing patient care. Future research should explore the barriers preventing some professionals from applying this effective methodology, despite its proven benefits and assess the impact of refresher courses on vertigo management, and evaluate the sustained benefits and economic viability of such interventions, including their potential to reduce unnecessary investigations and diagnostic delays. Also, it should also directly link improvements in diagnostic coding to patient-centered outcomes such as symptom resolution, quality of life, or healthcare utilization. Supplementary Information Supplementary Material 1. Supplementary Material 1.