Automated Videotopography for Dry Eye Diagnostics: Analytical Performance, Spatial Dynamics, and a Novel Multivariate Model

1. Introduction

Dry eye disease (DED) is a multifactorial condition characterised by a loss of homeostasis of the tear film, resulting in ocular symptoms and visual disturbances that affect between 5% and 50% of the global population [1,2]. As tear film instability is a core pathophysiological mechanism, its accurate and reliable quantification is critical for both diagnosis and management [1]. Historically, the clinical standard for this assessment has been the fluorescein tear film break-up time (FBUT). However, this method is encumbered by significant limitations: it is inherently subjective, leading to high inter- and intra-observer variability [1,3], and invasive, as the instillation of fluorescein dye alters the surface tension and volume of the tear film, potentially destabilising the very parameter under observation [4].

To address these measurement errors, automated non-invasive technologies have been developed. Systems such as the Oculus Keratograph 5M utilise videotopography, projecting a structured light pattern (e.g., Placido rings) onto the cornea. Automated software algorithms then analyse temporal distortions in the reflected image to quantify tear film break-up time (NIKBUT) and tear meniscus height (NIKTMH) without perturbing the ocular surface [3,5]. Despite the theoretical advantages of these objective systems, validation studies have consistently demonstrated that automated metrics are not interchangeable with manual clinical standards [6,7,8,9,10,11]. Large-scale investigations, such as the Dry Eye Assessment and Management (DREAM) study, reported weak correlations between Keratograph-derived metrics and traditional fluorescein break-up times [12]. Furthermore, a profound disconnect frequently exists between objective signs of instability and patient-reported symptoms, suggesting that single-metric thresholds may be insufficient for accurate diagnosis [1,12].

Current research gaps persist regarding the optimal utilisation of these automated systems. While global averages of tear stability are commonly reported, the spatial distribution of tear film break-up - whether it occurs centrally or paracentrally - may offer distinct diagnostic insights that global averages miss. Additionally, the reliance on isolated cut-off values for NIKBUT or NIKTMH often yields modest diagnostic sensitivity [4,7,13]. From a bioengineering perspective, this suggests that the diagnostic signal may not reside in a single parameter but rather requires a multivariate approach to bridge the sign-symptom gap.

Therefore, this study aimed to perform a comprehensive validation of an automated videotopography system. A prospective, repeated-measures design was employed to quantify the analytical performance (precision, reliability, and measurement error) of automated stability and volume metrics against manual standards. Furthermore, the study investigated the spatial dynamics of tear film instability and evaluated the clinical utility of a novel, device-native multivariate composite score (Objective-SRS). Principal findings indicate that while automated and manual methods are not interchangeable, a multivariate logistic model significantly outperforms individual indices in differentiating symptomatic from asymptomatic individuals, advocating for a computational approach to dry eye diagnostics.

2. Materials and Methods

2.1. Study Design and Participants

This prospective, single-centre validation study was conducted at the Centre for Contact Lens Research (CCLR), University of Waterloo, Canada, between February and April 2012. The study protocol adhered to the tenets of the Declaration of Helsinki and was approved by the University of Waterloo Office of Research Ethics (ORE #17216). All participants provided written informed consent prior to inclusion.

Participants aged 18 years or older were recruited from an existing patient database. Exclusion criteria included any active ocular pathology, use of ocular medications, contact lens wear within 24 hours of a study visit, and pregnancy or lactation. The study comprised three visits conducted on consecutive days at the same time of day (±10 minutes) to minimise diurnal variation. The starting eye and the initial NIKBUT illumination modality (white or infrared light) were randomly assigned to each participant. To minimise potential bias from tear-film disruption, clinical procedures were performed in a fixed sequence from least to most invasive. At each visit, participants first completed the Schein dry eye questionnaire, followed by non-invasive Keratograph 5M measurements (NIKTMH and NIKBUT). A mandatory washout period of at least 10 minutes was observed, during which participants completed the Ocular Surface Disease Index (OSDI) and McMonnies questionnaires. Subsequently, slit-lamp tear meniscus height (TMH) was measured. Following a second washout period of at least 10 minutes, the most invasive test, fluorescein break-up time (FBUT), was performed. All objective endpoints were captured three times per eye at each of the three visits.

2.2. Primary and Secondary Outcome Measures

Subjective Endpoint (Symptom Severity Score): To generate a robust, single measure of patient-reported symptoms, a composite score was derived. Raw scores from three validated questionnaires (OSDI, McMonnies, and Schein) were independently standardised into z-scores. The final Symptom Severity Score (SSS) for each participant was calculated as the arithmetic mean of these three z-scores.

For classification analyses, participants were dichotomised into “symptomatic” and “asymptomatic” groups based on the median split of the composite Symptom Severity Score. While it is acknowledged that dichotomising continuous variables can result in information loss and reduced statistical power, this approach was chosen to facilitate the calculation of standard diagnostic metrics (sensitivity, specificity) and to align with the binary nature of clinical decision-making (treat vs. no-treat). To ensure the robustness of this classification, the continuous relationship between the composite symptom score and the Objective-SRS was also assessed via correlation analysis.

Objective Endpoints (Clinical Tests):

• Fluorescein Break-Up Time (FBUT): The time in seconds from the last blink to the appearance of the first random dry spot, measured manually at the slit lamp.
• FBUT Spatial Location: The corneal segment (Superior, Inferior, Temporal, Nasal, or Central) where the first tear film break-up was detected.
• Tear Meniscus Height (TMH): The meniscus height in millimetres, measured manually at the slit lamp using a reticule scale.
• NIKBUT First: The time in seconds to the first detected distortion of the Placido rings, measured automatically by the Keratograph 5M.
• NIKBUT Average: The mean of all break-up times detected across the cornea during the measurement period.
• NIKTMH: The non-invasive tear meniscus height in millimetres, measured by the Keratograph 5M under infrared illumination.
• NIKBUT Break-up Area: The percentage of the total measurement area exhibiting tear film break-up during the recording period.
• NIKBUT Measurement Period: The total duration of the automated recording in seconds.

2.3. Development of the Objective Symptom Risk Score (Objective-SRS)

To evaluate the capability of automated metrics to predict symptom status, a multivariate Objective Symptom Risk Score (Objective-SRS) was developed using binary logistic regression. The model was trained on baseline (Visit 1) data to predict the binary symptom classification. Prior to modelling, all predictors were standardised (z-scored) to ensure coefficients were comparable and to aid model convergence. Collinearity diagnostics revealed high variance inflation factors (VIF > 70) between stability metrics. To mitigate this multicollinearity and ensure model stability, L2-penalised logistic regression (Ridge) was employed. The model incorporated five key Keratograph parameters: NIKBUT First, NIKBUT Average, NIKBUT Measurement Period, NIKBUT Break-up Area, and NIKTMH (Table 1).

The log-odds (z) for a participant being classified as symptomatic were calculated using Equation (1):

$$z={\beta }_0+\left({\beta }_1\times NIKBU{T}_{First}\right)-\left({\beta }_2\times NIKBU{T}_{Avg}\right)-\left({\beta }_3\times Measuremen{t}_{Period}\right)+\left({\beta }_4\times NIKBU{T}_{Area}\right)-({\beta }_5\times NIKTMH)$$

(1)

This log-odds value was transformed into a probability score (Objective-SRS) ranging from 0 to 1 using the standard logistic function shown in Equation (2):

$$Objective-SRS={\displaystyle \frac{1}{1+e^{-z}}}$$

(2)

The model coefficients, fixed from Visit 1 data, were applied to data from all three visits. Due to poor day-to-day reliability of the visit-specific scores (ICC = 0.426), the median of the three visit-specific scores was calculated for each participant. This median score served as the definitive Objective-SRS for all subsequent diagnostic and comparative analyses.

2.4. Spatial Allocation for Sectoral Analysis

For the automated NIKBUT, the ocular surface was mapped to a two-dimensional matrix of 192 areas, defined by 24 angular segments (s = 0-23) and eight concentric radial zones (f = 0-7). Radial zones were grouped into three categories: central (f=0), paracentral (f=1-3), and peripheral (f=4-7). Consequently, each matrix area was allocated to one of twelve sectors (e.g., central-superior, paracentral-nasal).

For the manual FBUT, a distinct allocation system was applied. The central corneal region was defined as a 3 mm diameter circle. The remaining surface was divided into four anatomical sectors (superior, inferior, nasal, temporal) using axes at 0°–180° and 90°–270°, with intermediate boundaries at 45° and 135°.

2.5. Statistical Analysis

All data processing and statistical analyses were performed using Python (v3.13.7) and relevant scientific libraries. A p-value < 0.05 was considered statistically significant. A hybrid imputation strategy was employed: missing NIKBUT values due to right-censoring (no break-up within 25 s) were imputed with the maximum value of 25 s using Tobit regression. All other missing data were handled using Multiple Imputation by Chained Equations (MICE). Five imputed datasets were combined using Rubin’s rules for the primary analysis, with sensitivity analyses conducted to confirm robustness.

Data from both eyes were pooled following paired t-tests that revealed no significant inter-eye differences (p > 0.05). Systematic changes across visits were assessed using Repeated Measures ANOVA. Group comparisons were performed using independent t-tests, Mann-Whitney U tests, or Chi-squared tests as appropriate. Analytical performance was quantified via: (1) Intra-session precision using the Coefficient of Variation (CV); (2) Inter-session reliability using the Intraclass Correlation Coefficient (ICC 3,k); (3) Measurement error using the Standard Error of Measurement (SEM) and Minimum Detectable Change (MDC95); (4) Method agreement using Bland-Altman analysis with random-effects; and (5) Spatial repeatability using Krippendorff’s Alpha.

Diagnostic accuracy was evaluated using Receiver Operating Characteristic (ROC) analysis to calculate the Area Under the Curve (AUC) with subject-level bootstrapping (resampling participants rather than records) to generate robust confidence intervals. Optimal cut-offs were determined by Youden’s J statistic. Clinical utility was assessed using Decision Curve Analysis (DCA) to quantify net benefit relative to “treat all” or “treat none” strategies, assuming a symptomatic prevalence of 51.4%.[14] Linear Mixed-Effects Models (LMM) assessed the fixed effects of Symptom Group, Visit, and Illumination on objective endpoints, with Subject ID as a random effect. A post-hoc power analysis was conducted based on observed effect sizes; sample size requirements for future validation were calculated to achieve 80% power () using participant-level aggregated data.

3. Results

3.1. Participant Characteristics and the Symptom–Sign Disconnect

The final analysis included 35 participants (69.4% soft contact lens wearers), classified as either symptomatic (n=18) or asymptomatic (n=17). While the groups were demographically similar, the symptomatic group exhibited significantly higher median symptom scores, confirming the validity of the classification method (Symptom Severity Score effect size, r=0.851). Preliminary analyses confirmed that data could be pooled across eyes (all p>0.47) and averaged across visits (all p>0.33).

The results revealed a significant disconnect between reported symptoms and objective signs. As detailed in Table 21, none of the individual objective clinical measures differed significantly between the two groups. The only metric to demonstrate a significant difference was the composite Objective-SRS (p=0.021), which yielded a large effect size (d=0.817). This discordance is further illustrated in Figure 1, where distributions for traditional metrics overlap substantially between groups.

Table 2. Baseline demographic and clinical characteristics. values are presented as mean ± standard deviation, median [Interquartile Range], or n (%).

Characteristic	Asymptomatic (n=17)	Symptomatic (n=18)	p-value
Age (years)	32.9 ± 11.5	28.1 ± 7.4	0.908
Sex (female)	10 (58.8%)	9 (50.0%)	0.854
Contact Lens Wearers	12 (71.4%)	14 (79.1%)	0.514
Schein Score	3.0 [2.0 - 4.0]	6.0 [3.2 - 8.0]	0.003
OSDI Score	4.0 [2.0 - 5.0]	11.0 [6.5 - 17.0]	0.001
McMonnies Score	2.0 [1.0 - 4.0]	9.0 [6.0 - 11.0]	0.001
Symptom Severity Score (SSS)	-0.544 [-0.764 - -0.359]	0.267 [0.176 - 0.807]	<0.001
Objective-SRS	0.544 ± 0.106	0.627 ± 0.097	0.021
FBUT (s)	9.69 ± 5.56 [6.83, 12.55]	8.53 ± 4.99 [6.05, 11.01]	0.525
Slit Lamp TMH (mm)	0.24 ± 0.05 [0.21, 0.27]	0.22 ± 0.04 [0.20, 0.24]	0.306
NIKBUT First BUT (s)	10.51 ± 4.52 [8.18, 12.83]	12.23 ± 4.58 [9.95, 14.51]	0.527
NIKBUT Average (s)	12.13 ± 4.19 [9.98, 14.29]	13.78 ± 4.02 [11.78, 15.78]	0.250
NIKBUT Area (%)	2.71 ± 1.02 [2.19, 3.24]	3.09 ± 1.66 [2.26, 3.91]	0.494
NIKBUT Max measuring time (s)	17.06 ± 4.06 [14.97, 19.15]	18.36 ± 3.59 [16.57, 20.14]	0.378
NIKTMH (mm)	0.34 ± 0.11 [0.28, 0.39]	0.31 ± 0.11 [0.26, 0.37]	0.507

Table 2. Baseline demographic and clinical characteristics. values are presented as mean ± standard deviation, median [Interquartile Range], or n (%).

Characteristic	Asymptomatic (n=17)	Symptomatic (n=18)	p-value
Age (years)	32.9 ± 11.5	28.1 ± 7.4	0.908
Sex (female)	10 (58.8%)	9 (50.0%)	0.854
Contact Lens Wearers	12 (71.4%)	14 (79.1%)	0.514
Schein Score	3.0 [2.0 - 4.0]	6.0 [3.2 - 8.0]	0.003
OSDI Score	4.0 [2.0 - 5.0]	11.0 [6.5 - 17.0]	0.001
McMonnies Score	2.0 [1.0 - 4.0]	9.0 [6.0 - 11.0]	0.001
Symptom Severity Score (SSS)	-0.544 [-0.764 - -0.359]	0.267 [0.176 - 0.807]	<0.001
Objective-SRS	0.544 ± 0.106	0.627 ± 0.097	0.021
FBUT (s)	9.69 ± 5.56 [6.83, 12.55]	8.53 ± 4.99 [6.05, 11.01]	0.525
Slit Lamp TMH (mm)	0.24 ± 0.05 [0.21, 0.27]	0.22 ± 0.04 [0.20, 0.24]	0.306
NIKBUT First BUT (s)	10.51 ± 4.52 [8.18, 12.83]	12.23 ± 4.58 [9.95, 14.51]	0.527
NIKBUT Average (s)	12.13 ± 4.19 [9.98, 14.29]	13.78 ± 4.02 [11.78, 15.78]	0.250
NIKBUT Area (%)	2.71 ± 1.02 [2.19, 3.24]	3.09 ± 1.66 [2.26, 3.91]	0.494
NIKBUT Max measuring time (s)	17.06 ± 4.06 [14.97, 19.15]	18.36 ± 3.59 [16.57, 20.14]	0.378
NIKTMH (mm)	0.34 ± 0.11 [0.28, 0.39]	0.31 ± 0.11 [0.26, 0.37]	0.507

3.2. System Characteristics and Linear Mixed-Effects Models

The influence of the illumination source (white vs. infrared) on NIKBUT metrics was minimal (Table S1, Supplementary Materials). Although pairwise comparisons suggested the NIKBUT Break-up Area was slightly larger under white light, the effect was inconsistent and small. A comprehensive Linear Mixed-Effects Model (LMM) confirmed that illumination had no significant main effect on any NIKBUT endpoint (p>0.05).

The LMM analysis further reinforced the central finding of a symptom–sign disconnect (Table S2, Supplementary Materials). With the exception of the composite Objective-SRS, Symptom Group was not a statistically significant predictor for any objective endpoint. Similarly, study visit and illumination type exhibited no significant fixed effects. Based on these outcomes, data from both illumination modalities were pooled for subsequent analyses to increase statistical power.

3.3. Analytical Validation

The analytical performance of the tests varied significantly (Table 3). Automated NIKTMH demonstrated excellent intra-session precision (CV = 8.8%) and the highest day-to-day reliability (ICC = 0.727). In contrast, NIKBUT metrics were highly imprecise (CV > 40%), while both the Objective-SRS and slit-lamp TMH demonstrated poor reliability (ICC approx. 0.4). Notably, the reliability of tear stability measures was consistently lower in the symptomatic group. From a clinical perspective, measurement error was substantial for some tests; the Minimum Detectable Change (MDC95) for FBUT was 9.3 seconds, indicating that only changes exceeding this threshold can be confidently attributed to clinical variation rather than measurement noise.

3.4. Method Comparison

Automated and manual methods were found not to be interchangeable. A correlation matrix revealed only weak correlations between automated and manual methods for both tear volume (NIKTMH vs. Slit Lamp TMH, r=0.33) and stability (Objective-SRS vs. FBUT, r=0.02) (Figure 2a).

Bland-Altman analysis confirmed this poor agreement (Figure 2b–d). For tear stability, NIKBUT measurements were consistently longer than FBUT, exhibiting exceptionally wide 95% limits of agreement and significant proportional bias. For tear volume, NIKTMH measured slightly higher than slit-lamp TMH, again with limits of agreement too wide for clinical substitution.

3.5. Diagnostic Accuracy

The composite Objective-SRS was the most accurate metric for differentiating between symptom groups. As shown in Figure 3 and Table 4, the Objective-SRS demonstrated the highest overall discriminatory power (AUC=0.768), achieving a sensitivity of 72.2% and specificity of 82.4% at its optimal data-driven cut-off. In contrast, all individual tests exhibited only fair discriminatory ability. When using established clinical cut-offs, the tests provided mixed performance, often characterised by high specificity (88.2%) but low sensitivity (27.8%), or vice versa.

3.6. Decision Curve Analysis (DCA)

DCA indicated that the Objective-SRS and manual Slit Lamp TMH offered the most significant clinical utility for decision-making (Figure 4). For tear stability tests, the Objective-SRS was the only metric that consistently provided a clear net benefit over the default strategy of treating all patients. For tear volume, the manual slit-lamp TMH demonstrated slightly superior clinical utility compared to its automated counterpart. The analysis suggested that these diagnostic tools are most useful for clinicians managing patients at moderate risk (i.e., at a 20–50% decision threshold), offering limited advantage in very low- or high-risk scenarios.

3.7. Spatial Dynamics

Analysis of tear film break-up revealed that NIKBUT and FBUT capture fundamentally different spatial phenomena (Figure 5 and Figure 6; detailed analysis in Tables S3 and S4, Supplementary Materials). For NIKBUT, break-up predominantly occurred in the paracentral zone, with a higher incidence in the symptomatic group (62.8%) compared to the asymptomatic group (53.1%). In contrast, FBUT events were heavily concentrated in the central zone for both groups, though significantly more so in asymptomatic (96.7%) than symptomatic (85.9%) patients.

Break-up time analysis also showed significant differences between locations for each method. For NIKBUT (Figure 5), central zones had the shortest break-up times, while paracentral zones (e.g., nasal, temporal) had the longest. For FBUT (Figure 6), symptomatic patients exhibited significantly shorter break-up times in the superior zone (-6.47 s, p<0.001) compared to asymptomatic patients. Despite these clear group-level patterns, the most critical finding was that the intra-subject repeatability of the break-up location was very poor (Krippendorff’s α=0.115–0.308), indicating that while spatial analysis provides mechanistic insight, the high variability of break-up location limits its reliability as a biomarker for individual patients.

3.8. Sensitivity and Power Analysis

A sensitivity analysis confirmed the robustness of the statistical models, demonstrating that the findings were not an artefact of the data imputation process. The post-hoc power analysis revealed that the Objective-SRS was a robust metric, successfully differentiating between symptomatic and asymptomatic participants with a large effect size (Cohen’s d=0.82); the current sample size provided 98% power to detect this difference. For future validation studies, a total of 48 participants would be required to replicate this effect with 80% power, while studies assessing reliability should aim for at least 50 participants.

4. Discussion

This validation study yields four principal inferences that refine the contemporary understanding of diagnostic testing in DED. First, it confirms a profound discordance between patient-reported symptom burden and objective clinical signs [1,18]. In the present dataset, no single objective endpoint differed significantly between symptom-defined groups, with the exception of the multivariable Objective-SRS, which showed significant separation. The mixed-effects models likewise demonstrated no systematic effect of symptom group, reinforcing the absence of a simple signs–symptoms mapping in this cohort. Second, the study quantifies analytical performance, demonstrating excellent precision and moderate-to-good day-to-day reliability for NIKTMH, contrasted with critically poor intra-session precision and symptom-dependent reliability for NIKBUT. Third, rigorous method-comparison statistics corroborate the non-interchangeability of automated and conventional measures. Finally, a multivariate, device-native composite (Objective-SRS) outperforms any single metric for differentiating symptomatic from asymptomatic participants, indicating that an informative diagnostic signal is distributed across parameters rather than concentrated in one measure [19].

The primary finding of a profound disconnect between signs and symptoms reinforces a well-established concept in DED, previously reported in large cohorts like the DREAM study [18,20,21]. By demonstrating this disconnect under rigorously controlled conditions, the results suggest it is a true biological feature of the disease rather than an artefact of testing variability. This aligns with the current TFOS DEWS III framework, which recognises that neurosensory abnormalities - alongside tear film instability and inflammation - are core aetiological drivers [1,22,23]. This sign–symptom disconnect may be explained by a neuropathic phenotype in the symptomatic group, where centrally mediated pain becomes decoupled from peripheral tear-film metrics [24,25]. Clinical data support this view, showing that patients with neuropathic ocular pain often exhibit severe symptoms with minimal objective signs [26]. A key clinical implication is that when a marked sign–symptom mismatch is observed, clinicians should consider targeted assessment for neuropathic pain, for instance, using a topical anaesthetic challenge [27].

Consequently, the lack of difference in individual tests like FBUT or NIKBUT should not be interpreted as instrument failure, but rather as an expected outcome of DED’s heterogeneous pathophysiology. The superior performance of the Objective-SRS is therefore critical. However, it is important to clarify the distinction between analytical validity and clinical utility. While the individual metrics showed poor analytical precision (high CV), the multivariate model demonstrated high diagnostic accuracy (AUC = 0.77). Clinically, the Objective-SRS is not proposed as a standalone diagnostic replacement for clinician judgement, but as a computational decision-support construct. By aggregating diffuse signals that are individually noisy, the algorithm provides a probability index that supports the clinician in managing the ambiguity of the symptom–sign disconnect. This motivates a paradigm shift: multivariable composite scores should be leveraged to broaden evaluation beyond the tear film to include neurosensory mechanisms when signs and symptoms are dissociated.

The analytical performance of volume- and stability-oriented tests diverged substantially, with direct consequences for clinical monitoring. While NIKTMH demonstrated excellent precision and the best day-to-day reliability, NIKBUT was highly imprecise, and its reliability was only moderate and dependent on symptom status. Critically, the large MDC95 for stability tests implies that substantial changes must be observed before they can be confidently attributed to biology rather than measurement noise. These performance differences lead to two key clinical inferences. First, FBUT is poorly suited for tracking incremental change; its large measurement error (MDC95 = 9.3 s) means a change from 4 to 8 seconds cannot be reliably distinguished from noise. In contrast, NIKTMH is a more plausible candidate for longitudinal surveillance of tear volume, provided that observed changes exceed its more modest MDC95 of 0.173 mm.

The finding that NIKBUT reliability was lower in symptomatic participants also has conceptual relevance. This could be explained within a neurosensory framework, where altered blink behaviour and sensory gain in symptomatic patients might inflate measurement variance [28]. This reframes high NIKBUT variability: rather than a simple technical limitation, it could function as an indirect clinical cue for underlying neurosensory dysregulation.

Method comparison analysis confirms the consensus that automated non-invasive indices are not interchangeable with their manual counterparts. Poor agreement was found between NIKBUT and FBUT, characterised by systematic bias (with NIKBUT values being longer) and exceptionally wide limits of agreement. This aligns with a large body of recent evidence demonstrating poor interchangeability across multiple devices [4,7,29]. Mechanistically, it is expected that the instillation of fluorescein for FBUT perturbs the tear film, artificially shortening break-up time values. Similarly, for tear volume, NIKTMH and manual slit-lamp TMH were not interchangeable due to wide limits of agreement. This finding must be contextualised by the fact that the clinical gold standard has since shifted to anterior segment OCT [30]. While early studies suggested the Keratograph underestimated TMH, recent work reports better agreement, likely due to software improvements [31]. This underscores a key challenge in the field: validation evidence for software-driven instruments can become outdated, requiring clinicians to remain vigilant about version-specific performance. Despite this, the findings confirm that automated and manual tests provide related but distinct information and should be used as complementary tools within a multimodal diagnostic strategy.

When using established clinical thresholds, individual tests proved to be poor diagnostic tools due to the classic trade-off between sensitivity and specificity. For example, an FBUT cut-off of <8 s had moderate sensitivity but poor specificity, while TMH and NIKTMH cut-offs achieved high specificity at the cost of very low sensitivity. ROC analysis confirmed the superiority of the composite score. While individual tests like FBUT and NIKBUT demonstrated only poor-to-fair discriminatory power (AUCs 0.56–0.63), the multivariate Objective-SRS was substantially more effective, achieving a good AUC of 0.768. Beyond statistical accuracy, Decision Curve Analysis (DCA) showed that the Objective-SRS was the only stability metric to provide a clear and consistent net benefit over the default strategies of treating all or no patients across a range of clinical scenarios.

Ultimately, these findings indicate that the diagnostic signal in DED is diffuse and cannot be captured by any single test. Meaningful diagnostic power emerges only when multiple parameters are integrated. This conclusion strongly resonates with the TFOS DEWS III diagnostic framework, which emphasises DED’s multifactorial nature and discourages reliance on single-metric cut-offs [1].

The spatial analysis suggests that NIKBUT and FBUT capture fundamentally different physiological phenomena. FBUT events were heavily concentrated in the central cornea, while NIKBUT events occurred predominantly in the paracentral regions. This supports the hypothesis that central FBUT may reflect aqueous deficiency patterns, while paracentral NIKBUT could indicate localised issues like poor wettability or altered lipid spreading, features consistent with evaporative dry eye [32,33]. This mechanistic distinction aligns with recent evidence suggesting that spatial mapping could help differentiate between DED subtypes [28,34]. However, despite these mechanistically plausible group-level findings, the clinical utility of spatial mapping is currently limited by extremely poor intra-subject repeatability. The location of the first tear film break-up appears to be a stochastic event, highly sensitive to micro-variations in blink completeness and local tear composition. Therefore, while spatial analysis is a valuable research tool for understanding disease heterogeneity, its high variability makes it unsuitable for routine clinical decision-making at the individual level.

This study has several notable strengths, including its prospective, repeated-measures design, which allowed for a robust quantification of intra- and inter-session variability and the calculation of clinically crucial metrics like the MDC95. The key innovation, however, is the development and validation of the Objective-SRS. By demonstrating that a multivariate composite index could capture variance missed by single tests, this work provides a template for extracting a more powerful signal from existing hardware.

The study’s findings must be interpreted within the context of the dataset’s age (2012). It is crucial to distinguish between findings that are algorithm-dependent versus those that are platform-invariant. Absolute values of NIKBUT and NIKTMH are subject to proprietary software changes and may differ in contemporary devices [21,31]. However, the platform-invariant findings—specifically the spatial distribution of break-up (central vs. paracentral), the magnitude of measurement noise (MDC95), and the fundamental symptom–sign disconnect - reflect physiological realities that persist regardless of software version. Furthermore, the Objective-SRS framework proposed here is intended to be device-agnostic in principle, serving as a valid engineering template even if specific coefficients require recalibration. Beyond the dataset age, other limitations exist. Methodologically, dichotomising participants into symptomatic and asymptomatic groups via a median split is a simplification of the continuous spectrum of DED, though it is necessary for binary classification modelling. Furthermore, the single-centre design and modest sample size may limit generalisability, although power analysis confirmed the study was well-powered (98%) to detect the large effect size of the Objective-SRS. Finally, while the neuropathic hypothesis provides a compelling framework, this study lacked direct measures of neurosensory function.

Future research should focus on three priorities: (1) External validation of the Objective-SRS in a multi-centre, demographically diverse cohort; (2) Contemporary re-validation benchmarking NIKTMH against anterior segment OCT; and (3) Mechanistic investigation incorporating neurosensory endpoints to directly test the neuropathic hypothesis.

5. Conclusions

This study highlights both the limitations and opportunities of modern non-invasive tear film assessment. It is concluded that the path forward for DED diagnostics does not lie in the search for a single perfect biomarker, but rather in the integration of multiple objective measures. The validation of the Objective-SRS demonstrates that intelligent integration of automated signals can bridge the sign–symptom disconnect, offering a superior method for aligning diagnostic output with the lived experience of patients and enhancing the precision of clinical decision-making.