Umeå University's logo

Background: Otitis media (OM) in children is a common infection in primary care, contributing to a significant global health and economic burden. In high-income countries, diagnostic inaccuracy leads to over-diagnosis of acute OM (AOM) and over-prescribing of antibiotics, which may contribute to antibiotic resistance.

Aim: To investigate the diagnostic accuracy and the influence of artificial intelligence (AI) in diagnosing OM among primary care physicians and medical students.

Method: A diagnostic accuracy study in which primary care physicians and medical students diagnosed AOM, OM with effusion (OME), and normal eardrums using 21 high-quality digital images, both without and with a fictitious AI support. We estimated the technological impact of the fictitious AI support.

Results: Overall diagnostic accuracy was 64% without, and 75% with AI support. The most experienced physicians reached 69% without, and 80% with AI; the least experienced 61% without, and 73% with AI; medical students reached 64% without, and 74% with AI. Accuracy for AOM was 77% without and 86% with AI, and for OME 46% without and 66% with AI. Mean diagnostic confidence increased significantly with AI support. The technological impact was 1.4. Automation bias was 1.2 overall, 0.9 for the most experienced and 1.2 for the least experienced physicians.

Conclusion: We report modest diagnostic accuracy for OM among primary care physicians and medical students. The fictitious AI support system improved both accuracy and diagnostic confidence and reduced over-diagnosis. The most experienced physicians achieved the highest accuracy, the less experienced were more often misled by the fictitious AI.

Background: Cardiovascular disease (CVD) is the leading cause of morbidity and mortality globally and in Sweden. Effective primary prevention requires not only clinical risk assessment but also person-centred communication to promote a healthy lifestyle. Nurses play a key role in structured preventive dialogues, such as those in the Västerbotten Intervention Programme (VIP), one of the world’s largest prevention programmes. However, little is known about how nurses experience and navigate risk communication in real-world primary care. This study aimed to explore nurses’ experiences with cardiovascular risk communication and lifestyle counselling during preventive dialogues in primary care.

Methods: Semistructured interviews were conducted with 11 nurses delivering VIP health dialogues across primary care centres in Västerbotten County between December 2024 and April 2025. The interviews were audio-recorded and transcribed verbatim. Data were analysed via qualitative content analysis, identifying meaning units, codes, categories, and an overarching theme. Reporting followed COREQ.

Results: The analysis generated two main categories. The first, Meeting the person professionally while taking a step back, captured how nurses described exploring each patient’s unique situation, needs, motivations, and attitudes. The second, Continuously adapting approaches in self-management support, captured nurses’ accounts of using flexible strategies, such as gradual goal setting, visual tools, and practical daily habit suggestions, to support patients in understanding and working with lifestyle change. Together, these categories formed the overarching theme Shifting between leading and following, illustrating how nurses experienced balancing individualised adaptation with professional guidance while navigating structural and contextual challenges.

Conclusions: Nurses described translating cardiovascular risk into person-centred, relational dialogue by flexibly combining structure, responsiveness, and motivational strategies. Understanding preventive dialogues as interactive, person-centred processes, supported by the sensitive use of visualisation tools, may inform the continued development of cardiovascular prevention in primary care and warrants further evaluation.

Diagnosing ear diseases from images of the tympanic membrane has emerged as a promising application for deep learning-based classification models. While several studies report high diagnostic accuracy, these models often struggle to generalize to external data that does not match the image characteristics of the training data. This study aims to evaluate methods for enhancing model generalizability. We trained artificial neural networks on tympanic membrane images from five datasets with a large heterogeneity between them. The datasets consisted of images from different countries, captured with different otoscope models, and with different image pre-processing applied at the source (e.g., colour normalization, cropping, and resolution). Artificial neural networks for binary (normal/abnormal) classification were trained on three datasets, and the models were evaluated on two. We explored several strategies to improve generalizability, including image pre-processing, data augmentation, transfer learning, and different network architectures. The effects of each investigated factor and their interactions were determined through a full factorial experimental design and linear mixed-effects model. Our best classification network achieved a mean accuracy of 83% (sensitivity 71%, specificity 91%) on external datasets. In contrast, the baseline trained without an optimized training method and using non-processed datasets yielded an accuracy of 53% (sensitivity 83%, specificity 34%). Our findings highlight how a better choice of network, training and augmentation strategies, and data verification improve model generalizability.

Introduction: Understaffing and dependence on temporary physicians known as locums are common in primary healthcare in Sweden. In this study, we investigate the impact of locum staffing on the quality of antibiotic prescribing for respiratory tract infections. We used two neighbouring cities in northern Sweden with large differences in staffing conditions: Skellefteå, with a 10-fold higher reliance on locum physicians compared to Umeå, which has low locum staffing.

Methods: We used data extracted from patient records for all visits, from the four largest healthcare centres in each of the cities Umeå and Skellefteå, for respiratory tract infections, including the ICD-10 diagnosis code and whether antibiotic treatment was prescribed or not. Data on the contracted hours of locum staff for each healthcare centre were also obtained. Statistical analyses were performed with Spearman correlation, Chi-square and Poisson tests. Logistic regression was used to investigate whether prescribing patterns were affected by the level of relay physicians.

Results: We analysed antibiotic prescriptions for respiratory tract infections based on data from 11,228 patient visits. Regardless of the statistical method used, we found a statistically significant difference, with higher rates of antibiotic prescribing for diagnoses where antibiotics generally are not recommended or where they should be prescribed restrictively. For pneumonia, where antibiotics should normally be prescribed, no difference was seen. The odds of antibiotic prescription for acute upper respiratory tract infection in Skellefteå were 2.5 times higher than in Umeå, with a positive correlation between locum staffing and antibiotic prescriptions for diagnoses where antibiotics should be used restrictively.

Conclusions: Our findings show that a high level of locum staffing and thus insufficient staffing with permanent General practitioners leads to higher rates of inappropriate antibiotic prescribing. This is shown both at the level of individual primary care healthcare centres and at the community level.

Otitis Media (OM) is a prevalent condition in children that can lead to hearing impairment and significant healthcare costs. Inaccuracy in primary care and equipment cost in developing countries are concerning issues in OM diagnostics. Acoustic Reflectometry (AR) offers a low-cost, non-invasive diagnostic alternative, though it has fallen short on accuracy in previous studies. The primary aim of this study was to establish a computational simulation and an experimental model able to reproduce AR performed on human individuals to enable further research and accuracy improvement. The secondary aim was to perform a sensitivity analysis on AR instrument user error. Simulations and experiments were validated against measurements from human individuals with OM and normal ears, respectively. The results reveal that the simulation sufficiently reproduces human AR measurements and distinguishes an ear with OM from a healthy ear. The experiment delivered satisfying measurements on OM but underperformed in a normal ear scenario. The simulations and experiments overpredicted sound reflection in OM. The sensitivity study showed promising robustness of AR, concluding that computational simulation is a viable tool and complement to an experimental approach in research of AR. Future efforts should focus on paediatric models and partially filled middle ear simulations to promote clinical relevance.

Background: Otomicroscopy and pneumatic methods are superior to otoscopy alone in diagnosing acute otitis media (AOM). There is a lack of knowledge regarding the use of different diagnostic methods for AOM in primary health care in Sweden and Norway.

Methods: This cross-sectional study included a questionnaire completed by general practitioners (GPs) and specialist trainees (STs/residents/registrars) working in primary care in Sweden and Norway. Multivariable binary logistic regressions were performed to evaluate the use of diagnostic methods and written advice adjusted for educational level, sex and country.

Results: Otoscopy was the most frequently used method. Sweden had greater access to the more accurate diagnostic methods. In Norway, the following methods were used to a lesser extent: pneumatic otoscopy, adjusted OR 0.15 (95% CI 0.10–0.23; p < .001), otomicroscopy, adjusted OR 0.013 (95% CI 0.070–0.027; p < .001), pneumatic otomicroscopy, adjusted OR 0.028 (95% CI 0.010–0.078; p < .001) and tympanometry, adjusted OR 0.31 (95% CI 0.21–0.45; p < .001). Written advice was used to a greater extent in Norway, adjusted OR 4.5 (95% CI 3.1–6.7; p < .001). The STs used pneumatic otoscopy and pneumatic otomicroscopy to a lesser extent, adjusted OR 0.65 (95% CI 0.45–0.93; p = .019) and 0.63 (95% CI 0.43–0.92; p = .016).

Conclusions: Swedish physicians both used and had greater access to the significantly better diagnostic methods compared with Norwegian physicians while the opposite applied to the use of written information. The GPs used pneumatic otoscopy and pneumatic otomicroscopy to a greater extent than STs. Compared with 2012, the Swedish physicians now more frequently used pneumatic otoscopy.

Background: Otitis media includes several common inflammatory conditions of the middle ear that can have severe complications if left untreated. Correctly identifying otitis media can be difficult and a screening system supported by machine learning would be valuable for this prevalent disease. This study investigated the performance of a convolutional neural network in screening for otitis media using digital otoscopic images labelled by an expert panel.

Methods: Five experienced otologists diagnosed 347 tympanic membrane images captured with a digital otoscope. Images with a majority expert diagnosis (n = 273) were categorized into three screening groups Normal, Pathological and Wax, and the same images were used for training and testing of the convolutional neural network. Expert panel diagnoses were compared to the convolutional neural network classification. Different approaches to the convolutional neural network were tested to identify the best performing model.

Results: Overall accuracy of the convolutional neural network was above 0.9 in all except one approach. Sensitivity to finding ears with wax or pathology was above 93% in all cases and specificity was 100%. Adding more images to train the convolutional neural network had no positive impact on the results. Modifications such as normalization of datasets and image augmentation enhanced the performance in some instances.

Conclusions: A machine learning approach could be used on digital otoscopic images to accurately screen for otitis media.

Background: There is a lack of hearing health care globally, and tele-audiology and mobile technologies have been proposed as important strategies to reduce the shortfall. Objectives: To investigate the accuracy and reliability of smartphone self-test audiometry in adults, in community clinics in low-income settings.

Methods: A prospective, intra-individual, repeated measurements design was used. Sixty-three adult participants (mean age 52 years, range 20-88 years) were recruited from ENT and primary health care clinics in a low-income community in Tshwane, South Africa. Air conduction hearing thresholds for octave frequencies 0.5 to 8 kHz collected with the smartphone self-test in non-sound treated environments were compared to those obtained by reference audiometry.

Results: The overall mean difference between threshold seeking methods (ie, smartphone thresholds subtracted from reference) was -2.2 dB HL (n = 467 thresholds, P = 0.00). Agreement was within 10 dB HL for 80.1% (n = 467 thresholds) of all threshold comparisons. Sensitivity for detection hearing loss >40 dB HL in one ear was 90.6% (n = 84 ears), and specificity 94.2% (n = 84 ears).

Conclusion: Smartphone self-test audiometry can provide accurate and reliable air conduction hearing thresholds for adults in community clinics in low-income settings.

Objective: Video-otoscopy is rapidly developing as a new method to diagnose common ear disease and can be performed by trained health care facilitators as well as by clinicians. This study aimed to compare remote asynchronous assessments of video-otoscopy with otoscopy performed by a general practitioner. Method: Children, aged 2-16 years, attending a health center in Johannesburg, South Africa, were examined. An otologist performed otomicroscopy and a general practitioner performed otoscopy. Video-otoscopy was performed by a health care facilitator and video sequences were stored on a server for assessment by the same general practitioner 4 and 8 weeks later. At all examinations, a diagnosis was set and the tympanic membrane appearance was graded using the OMgrade-scale. The otologist's otomicroscopic diagnosis was set as reference standard to compare the accuracy of the two otoscopic methods. Results: Diagnostic agreement between otologist's otomicroscopic examination and the general practitioner's otoscopic examination was substantial (kappa 0.66). Agreement between onsite otomicroscopy and the general practitioners asynchronous video assessments were also substantial (kappa 0.70 and 0.80). Conclusion: Video-otoscopy performed by a health care facilitator and assessed asynchronously by a general practitioner had similar or better accuracy compared to face-to-face otoscopy performed by a general practitioner.

Background Acute otitis media (AOM) is a common disease in children and is causing great discomfort and disability worldwide but many areas are underserved regarding skilled professional. Tele-otology offers a promising technique to provide ear health globally. Diagnostic accuracy of AOM has regardless of method been found to be low. Grading the severity of AOM my offer a guide in decision on antibiotic treatment, however grading systems need improvement.

Aim To describe and evaluate imaging of the tympanic membrane (TM), develop an image based grading scale for AOM and to study the characteristics and the course of acute otitis media (AOM) episodes in children with the use of telemedicine techniques.

Method This thesis is based on two study populations, 63 children attending with othalgia at four primary health care centers in rural Sweden (papers I, II, IV) and 140 children attending a health clinic from a township in Johannesburg, South Africa, (paper III).

Paper I: Image quality of endoscopic imaging of TM´s, from the Swedish study was assessed by an otologist and two general practitioners together with an evaluation of important characteristics of assessing TM appearance.

Paper II: In development and validation of an image-based grading scale of AOM two expert panels of otologist’s evaluated the proposed grading scale stepwise and in a test and retest validation process.

Paper III: A test of the scale in a clinical situation was set up, an otologist oto-microscopically examined children and used the grading scale, and his diagnoses were set as gold standard. A specially trained ear and hearing facilitator then recorded videos of the TM using video-otoscopy. Videos were remotely assessed by the same otologist and by a general practitioner twice; 4 and 8 weeks after the otologist’s on-site grading.

Paper IV: Children with othalgia were followed with assessments of their symptoms and signs over a period of 3 months. An assessment group of two general practitioners and one otologist evaluated TM images, tympanograms and recorded symptoms and make a diagnose.

Results The results from paper I show that image quality was good and the position and transparency of the TM was found to be the most important characteristics when assessing TM. In paper II the new grading scale (OMGRADE) was developed and validated. The image-based scale focuses on the position and transparency of the TM. The results from paper III showed that the OMGRADE scale could discriminate the normal ear as well as ears with otitis media with effusion (OME) in an unselected pediatric population. Paper IV showed that the bilateral AOM had more severe symptoms. The children with chagrinated TM’s took the longest time to resolve regarding TM appearance and tympanograms. Furthermore, symptoms resolved quicker than TM changes and tympanograms during the first week.

Conclusions TM images or video recordings taken by a trained nurse or facilitator are sufficient for remote evaluation. The new grading scale of TM appearance is valid and reliable and may function as a diagnostic guide together with evaluation of middle ear effusion. TM appearance may be of importance in grading the severity of an AOM episode.