This exploratory mixed methods evaluation was cross-sectional, survey-based, and interview-based, designed to characterize clinician experience across 4 ASP workflows rather than test predefined hypotheses. The project ran from October 21 to December 6, 2024, at Moorfields Eye Hospital (MEH; City Road site, United Kingdom) eye casualty and outpatient cornea clinics. As an exploratory, single-center quality improvement evaluation, this project is intended to characterize clinician experience and generate hypotheses for future research, rather than provide definitive or generalizable comparisons between ASP workflows.

Four standardized ASP workflows, each including distinct categories, were evaluated in this project in the eye casualty and corneal services. Workflows were defined and categorized, according to a recent scoping review, into smartphone-based devices (with or without attachments) and slit-lamp–based devices (slit-lamp cameras, conventional slit-lamps equipped with a digital camera, and slit-lamps equipped with a smartphone) [6]. Each device was integrated into a standard clinic workflow. A workflow was defined as the following standard of care: a patient was invited from the waiting room to an examination cubicle (estimated 12 m²), an appropriate history and examination were conducted after obtaining appropriate consent, the room light was turned off to maximize slit-lamp examination findings or set to ambient lighting (estimated 4000 K) to maximize smartphone-based examinations, and the specific ASP workflow of interest was assessed according to the manufacturer’s photo-capture instructions. These are described in further detail below.

The smartphone-based workflows comprised the Magnifier app (Apple Inc; Figure 1A), an iOS app running on an iPhone 13 Pro (Apple Inc) that isolates the ultrawide (macrocapable) lens to provide steady zoom and fixed focal length, and the QuikVue (VisuScience Meditech Co Ltd; Figure 1B), a macro lens attachment with built-in illumination and cobalt-blue filter, providing 10× magnification at a fixed focal length when coupled to a compatible smartphone [7,8].

The slit-lamp–based workflows comprised the NexYZ (Celestron LLC; Figure 1C), a universal smartphone adapter mounted to the slit-lamp eyepiece to enable image capture through slit-lamp optics, and the Zeiss Slit Lamp Imaging Solution (ZSLIS, Carl Zeiss AG; Figure 1D), an integrated slit-lamp camera system producing high-resolution images via a dedicated optical and camera assembly [9,10]. The NexYZ workflow was assessed by coupling a smartphone to the adapter before mounting onto the slit-lamp (Haag-Streit BM 900), thereby using the optics of both the smartphone and the slit-lamp (diffuse illumination settings: 10× magnification, illumination angle 45°, 8×14 mm beam [height × width], and low illumination intensity with a diffused light source). The ZSLIS workflow was the only one that did not require a smartphone for image capture.

In the case of smartphone-based workflows or slit-lamps equipped with a smartphone, a standardized iPhone 13 Pro (owned by a member of the project team with no ancillary accessories, such as a case) was used with the standard camera app native to iOS or the Magnifier app. The settings across ASP workflows changed only when applied to the Magnifier app, which isolates the ultrawide camera (12-megapixel, 13 mm lens, aperture f/1.8, and shutter speed 1/100 s). For all other workflows, the smartphone Camera app settings were set as follows, using the main telephoto camera (12-megapixel, 26 mm lens, aperture f/2.8, and shutter speed 1/100 s): fixed zoom (1×), fixed brightness and exposure at the automatic midpoint, no flash, auto-focus, and image-size exportation (maximum resolution, 4032×3024) [11].

Throughout this manuscript, references to a named ASP device (eg, the NexYZ) denote the complete clinical workflow in which it was evaluated. This includes, if applicable, the coupled smartphone, slit-lamp optics, room lighting, capture software with associated settings, and the image-handling pathway, rather than the isolated device. Outcomes reported reflect this full workflow and not the hardware component alone. For the purposes of readability, a named ASP device (eg, the NexYZ) may be used as shorthand for the complete workflow, not the isolated component.

This project was registered with and approved by the MEH Clinical Audit Department as a quality-improvement project (QIP; project number 1606). Formal National Health Service (NHS) Research Ethics Committee review was not sought because the project was classified locally as a QIP rather than research. This classification reflected that the project evaluated clinicians’ experience of ASP workflows already available or in use in standard clinical practice, with the primary purpose of informing local service improvement, workflow integration, documentation, and governance. The project did not allocate participants to an intervention, alter clinical care, or test a predefined hypothesis of clinical efficacy or diagnostic accuracy. This approach is consistent with UK Health Research Authority guidance distinguishing research from audit or service evaluation [12].

All clinician participants participated voluntarily and provided informed consent before survey participation and/or interview audio recording. For the illustrative clinical case images, patient consent was obtained in accordance with the MEH Patient Agreement to Media Recordings. This meant all patient-identifiable images were handled in line with the Data Protection Act 2018, the UK General Data Protection Regulation (GDPR), NHS England Bring Your Own Device (BYOD) guidance, and local governance policies [13,14]. Clinical images were captured on an approved iPhone 13 Pro using airplane mode to prevent transfer of images via cellular or unsecure networks. Captured images were immediately transferred to a secure NHS file server via a USB-linked file transfer to a clinical computer. Images were deleted immediately following transfer from the iPhone 13 Pro: no images were stored in the device’s general photo library or in an app-specific library, or transmitted via personal messaging or unsecure networks. This is all per the NHS England BYOD guidance and in keeping with local data protection policies. Survey and interview data were anonymized and deidentified before analysis. No participants received compensation for project participation.

Corneal fellows and ophthalmology residents at MEH, City Road, were recruited to complete a survey and voluntary semistructured interviews. The recruitment process involved a project coordinator (VN) rotating through morning and afternoon sessions in the eye casualty and outpatient corneal clinics and approaching residents and fellows in these clinics about the QIP, with the available workflows present (excluding the ZSLIS); a total of 6 eye casualty sessions and 5 corneal clinics were visited during the period of October 21 to December 6, 2024.

All ophthalmology residents receive induction training for the ZSLIS prior to commencing a placement at MEH; therefore, familiarity with this system was assumed. All project participants had prior experience with smartphone-based ASP (defined as a general familiarity with smartphone photography for clinical ophthalmic assessment rather than the specific workflows tested), ensuring baseline familiarity. Prior experience with a tested ASP workflow was not an exclusion criterion, since the clinician’s perspective was still valuable information. Participants were further engaged through a hands-on familiarity exercise, lasting approximately 5 minutes and including practice in image capture and free exploration of each workflow’s features, prior to survey and interview administration.

To provide illustrative images and contextualize each ASP workflow within a diagnostic pathway, a single clinical case was used. A patient with corneal pathology attending the eye casualty was recruited with informed consent in accordance with the MEH Patient Agreement to Media Recordings. Images were captured by an ophthalmology resident (YS-B) using each of the 4 ASP workflows, including the same smartphone (iPhone 13 Pro) where applicable, during a single clinical encounter. Image acquisition was performed in either ambient lighting (estimated 4000 K) for smartphone-based examinations, except for the NexYZ which used slit-lamp optics, or with the room lights turned off for slit-lamp–based examinations. Two clinical examination cubicles were used; the ZSLIS required a separate cubicle as it used an integrated imaging system with a dedicated computer. Files from the ZSLIS were transferred to a secure NHS email account and stored on a secure NHS file server. For all other smartphone-based assessments, the native iOS camera app (iPhone 13 Pro) was used, except for the Magnifier workflow. Magnifier and camera settings were applied as described previously, and cubicle lighting and patient positioning were kept consistent between the 2 environments.

Participants were introduced to the project aims and given a short tutorial on each ASP workflow to standardize knowledge among clinicians. Surveys and interviews were designed to maximize user reflections on each ASP workflow and were not intended for formal hypothesis generation or statistical comparison. Each interview concluded with a discussion of differential diagnoses for a single clinical case, presented as the following vignette: a young male contact lens wearer presenting to eye casualty with a 24-hour history of eye pain and a corneal opacity on examination. Further discussion around global image characteristics across all 4 ASP workflows, including perceived functionality and costs, was held during interviews.

Formal sample size and power calculations were not conducted, reflecting the project’s exploratory design. Instead, target sample sizes were informed by prior literature and the practical limitations of recruiting within a niche specialty cohort. Twelve interviews were deemed sufficient for achieving thematic saturation, beyond which no new codes, subthemes, or themes are generated [15]. A sufficient benchmark for survey responses was deemed based on the limited number of residents at a single tertiary eye center, ranging between 20 and 30, and previous survey-based studies, among ophthalmology residents, ranging between 16 and 26 [16,17].

The survey underwent dynamic iterative review (VN, YS-B, and RR) to optimize its design, alongside the adaptation of previously validated surveys: the System Usability Scale (SUS), a simple 5-point Likert scale (eg, “strongly disagree,” “disagree,” “neither agree nor disagree,” “agree,” and “strongly agree”) used to measure subjective assessments of usability, and the mHealth App Usability Questionnaire (MAUQ), a questionnaire developed from existing validated surveys to assess the usability of a novel health care smartphone app [18,19]. The final survey (Multimedia Appendix 1) used for our project was nonvalidated, but its components have been previously used in the ophthalmology field [20]. There is evidence to suggest acceptable reliability across a range of usability scales, from single to multi-item measures, suggesting that the underlying constructs are robustly captured even when item sets are modified for contextual fit [21].

The statements were adapted from the MAUQ to focus on the image-capture workflow rather than a smartphone app, assessing 3 domains: ease of use, interface and satisfaction, and usefulness. Statements 1 to 2 of the survey covered “ease of use,” 3 to 5 addressed “interface and satisfaction,” and 6 to 7 focused on “usefulness.” The SUS and MAUQ were not applied as stand-alone validated instruments in their original scored form; instead, only selected items served as a conceptual source for usability framing and descriptive statistics. The use of single-item usability measures has been demonstrated to be equally reliable as more substantially itemed questionnaires [21,22].

A 5-point Likert scale, adapted from the SUS, was included to provide a numerical measure for each subjective question, ranging from 1 to 5. An open-text box allowed for additional reflective comments from survey participants. The same survey was presented for each workflow. Price ranges were provided for each workflow to inform the response to statement 7, referring to cost-effectiveness. The survey also collected participants’ staff grades while maintaining full anonymity. Finally, a question at the end was provided to invite participants to opt into a semistructured interview for further discussion and reflection on ASP workflows.

Surveys were disseminated within a single center. Participants were introduced to the project aims and given a brief tutorial on workflow operations to ensure consistent understanding, alongside a familiarity exercise. The evaluation focused on user experiences, both prior and novel, to provide a qualitative, quality-improvement perspective. Each workflow was trialed synchronously before survey completion, supporting experiential comparisons. Following survey completion, one-on-one interviews were conducted face-to-face or virtually via Zoom using an interview guide (Multimedia Appendix 2), with verbal consent obtained for audio recording. Zoom interviews were offered to accommodate clinicians’ schedules rather than to selectively recruit technology-savvy participants.

To extend the analysis beyond usability and user preferences, the potential clinical applicability of ASP workflows was explored. This is particularly important given that many BYOD and ASP systems are increasingly designed not only for image acquisition but also to support clinical decision-making. Therefore, our project question in this regard was to ascertain what the clinical applications of these ASP workflows were with respect to clinical decision-making, specifically in informing diagnostic and monitoring decisions.

At the end of each interview, respondents were presented with a brief clinical case involving microbial keratitis, a common ophthalmic condition with potentially sight-threatening consequences if misdiagnosed or inadequately monitored. The condition was deliberately selected as a straightforward and common clinical scenario in which image quality may directly influence diagnostic confidence and longitudinal assessment. Participants were shown 4 unlabeled images of the same patient’s eye, each captured using a different ASP workflow (Figure 2). Due to the visual nature of the ophthalmology specialty, diagnostic images were presented on printed A4 paper, and no additional clinical information or prompting was provided beyond the vignette.

Clinicians were asked to comment on image characteristics, identify relevant pathology, compare image quality between workflows, and match each image to its perceived source before answers were revealed. This exercise was designed to assess whether images obtained using different ASP workflows could provide sufficient information to support clinically meaningful decisions. With respect to the clinical utility scoring, participants were asked whether each image was adequate to inform a diagnostic or monitoring decision independently, or whether a conventional slit-lamp examination would still be required. This approach allowed the evaluation of the potential role of ASP workflows in supporting real-world clinical pathways beyond image capture alone.

The survey results were collated and analyzed in Microsoft Excel, and interviews were transcribed using Otter.ai and subsequently reviewed for accuracy by the authors (VN and YS-B). Consistent with the exploratory aims of this project, survey data were summarized descriptively; mean (SD) Likert-scale responses were calculated for each item, used to contextualize qualitative findings, and presented in a table. Additional comments from free-text boxes were thematically analyzed alongside interview transcripts using NVivo (Lumivero).

The use of means and SDs to summarize Likert-scale responses is a widely debated but commonly accepted practice in exploratory mixed methods studies [23,24]. Given the exploratory and quality improvement framing of our project, mean Likert-scale scores were intended as descriptive summaries to contextualize qualitative findings rather than as inferential statistics. These values summarize clinician perceptions of the complete imaging workflows and should be read as descriptive only. They are influenced by confounding factors, such as smartphone familiarity, capture software, and prior experience with BYOD governance, and are not intended to support statistical comparisons between ASP workflows. This is particularly relevant for the “intention to use again” domain, which is shaped by a clinician’s existing familiarity with and acceptance of ASP rather than the imaging workflow alone. Formal hypothesis testing with statistical analysis was outside the scope of this evaluation, given the exploratory design and sample size.

The authorship team undertook a reflexive exercise prior to interviews, in which personal assumptions and positionalities were explicitly acknowledged as an active part of the analytical process; a predominantly constructivist research paradigm was adopted for the mixed methods project. All transcripts and comments were coded by a single author (VN); codes were iteratively developed into candidate subthemes, with those sharing analytical constructs consolidated into themes. This coding was then critically reviewed and interrogated by 2 authors (YS-B and RR) acting as “critical friends,” using an online mind-mapping process (Canva) to challenge and refine the developing themes rather than establishing intercoder agreement. This generative process of codes, subthemes, and themes reflects the interpretive flexibility of reflexive thematic analysis, wherein analytic constructs are understood to be author-generated and subject to human bias, rather than an inherent property of the data [25,26]. The thematic analysis reflects the mixed methods nature of our project, where interpretation of data is subjective, adopting a constructivist research paradigm.

The quantitative and qualitative components were analyzed and reported separately in the Results section. Likert-scale responses were summarized descriptively and were not incorporated into the thematic analysis. The only exception was that free-text survey comments were pooled with the interview transcripts and coded as a single qualitative dataset, so that both sources contributed to the same subthemes and themes. Descriptive Likert-scale ratings and qualitative themes were considered together at the discussion stage, where consistency between them supported an interpretation, and discrepancies were retained as reflecting variation in clinician experience.

A total of 35 residents and fellows were approached over a 4-week period, with 25 survey responses collected: 17 (68%) ophthalmology residents and 8 (32%) corneal fellows; 10 participants declined to participate due to timing constraints. Of this cohort, 16 expressed interest in an interview, and 12 interviews were completed (25% drop-out rate). The mean interview duration was 18.1 (SD 10.8) minutes.

The survey data were first charted by collating Likert scale responses for each ASP workflow. Mean Likert scale response scores were calculated for each question and imaging workflow and are presented in Table 1. Clinician perceptions favored the Magnifier workflow, which received the highest ratings for usability, cost-effectiveness, and clinical utility (19/25, 76%, 23/25, 92%, and 18/25, 72% strongly agreed, respectively), with consistently favorable ratings across the evaluated domains.

Interview respondents were asked about their preferred ASP workflow in the context of clinical utility, assessed through a review of captured images from a clinical case of microbial keratitis. The respondents’ preferences also included a subjective summation of usability aspects, as explored through thematic analysis and surveys. It was revealed that the majority preferred the Magnifier workflow. Table 2 shows overall preferences in a clinical utility context, though these findings reflect clinician-perceived clinical utility based on a single illustrative case.

When presented with images of a corneal infiltrate from the clinical case of microbial keratitis, all respondents identified the correct clinical diagnosis and localized it to the appropriate region of the cornea. Across workflows, variation was observed in global image characteristics such as field of view, depth of field, resolution, illumination, and the presence of reflections. For example, the Magnifier image encompassed the entire globe, allowing appreciation of conjunctival injection in addition to a faint outline of the infiltrate. The QuikVue image demonstrated higher magnification, but was limited by prominent reflections, reducing structural visibility. The NexYZ image, while also highly magnified and lacking conjunctival context, conveyed greater information about the depth of the lesion. The ZSLIS image offered the highest resolution, with fine adnexal detail, though uneven illumination introduced shadowing at the lesion site.

When explored in interviews, clinician preferences reflected these differing characteristics: the NexYZ workflow was selected as the most informative for diagnostic purposes (n=5, 42%), while the Magnifier workflow was most frequently identified as the second most useful (n=4, 33%). It should be emphasized that these results highlight a preference for the ASP workflow in the context of producing a diagnostically informative image during interviews, rather than an objective diagnostic accuracy assessment.