A smartphone-based blink training application for alleviating dry eye signs and symptoms

Abstract

Dry eye disease (DED) is exacerbated by prolonged screen use, partly due to reduced blink frequency and a higher percentage of incomplete blinks (IB). In this prospective, randomized study, we assessed whether a smartphone‑based blink‑training application alleviates the signs and symptoms of DED. Forty participants were randomly assigned to either an intervention group that used the application or a control group without intervention, and they were followed for 30 days. Primary outcomes were Ocular Surface Disease Index (OSDI) score, blink rate (BR), percentage of IB, Ocular Protection Index (OPI), and tear‑film stability. After 30 days, the intervention group showed significant improvements in blink behavior and ocular surface parameters, whereas the control group showed no meaningful changes. Generalized estimating equations indicated that training duration, OPI, BR, and sex were independently associated with changes in the OSDI scores. These findings suggested that a simple, app‑based behavioral intervention yields measurable benefits for DED management and provided a scalable, cost‑effective strategy for ocular health preservation in digital environments.

Introduction

Computer vision syndrome (CVS), first described in 2002, refers to ocular discomfort associated with video display terminal (VDT) use, including fatigue, burning, irritation, and sensations of dryness or astringency¹. Dry eye disease (DED) is now recognized as a principal driver of CVS^2,3. Prolonged VDT use can, in turn, exacerbate DED by reducing blink rate (BR) and increasing the percentage of incomplete blinks (IB), creating a self‑perpetuating cycle^4,5.

Recent clinical work has highlighted the influence of blink behavior on tear-film dynamics in DED⁶. Larger IB percentages and lower BRs have been linked to more-severe symptoms⁷, and IB frequency is significantly higher in individuals with DED than in those without the condition⁸. Sustained visual concentration suppresses BRs and lengthens ocular-surface exposure time⁹. Moreover, IBs predominate in the blink patterns of professionals who rely heavily on electronic displays^10,11.

Optimizing blink behavior may therefore relieve DED symptoms¹². The TFOS DEWS II Management and Therapy Report recommends “frequent breaks and blinking-awareness training” for users who engage in prolonged VDT work¹³. In 2015, Nosch et al. introduced Blink Blink, a desktop blink-training program that reported improvements only in the Ocular Surface Disease Index (OSDI)¹⁴. Subsequent work by Ashiwini et al. evaluated a similar intervention but omitted objective ocular-surface metrics¹⁵.

Smartphone ownership has soared in recent years, increasing daily VDT exposure, and contemporary studies have linked smartphone use to ocular discomfort and tear-film instability^16–18. The latest TFOS DEWS III Management and Therapy Report states that optimizing blink frequency and completeness is one of the simplest yet most difficult methods to implement in managing DED¹⁹. Mobile applications that reshape blink behavior could therefore represent an accessible adjunct to DED management. To address limitations of previous research, we incorporated both subjective and objective ocular surface outcomes, detailed blink metrics, and a 30-day follow-up to evaluate the therapeutic potential of smartphone‑based blink training. We hypothesized that app-guided blink training would improve blink patterns and ameliorate the signs and symptoms of DED in participants who used smartphones daily.

Results

Demographics

The demographic and clinical characteristics of the two groups were presented in Table 1. The baseline assessment revealed no statistically significant differences between the groups across all evaluated parameters. No adverse events were reported during the study period.

Table 1.

Baseline demographic characteristics and DED parameters of the two groups

Parameters	Eye	Control Group (Mean ± SD / Median [Q1-Q3])	Experimental Group (Mean ± SD / Median [Q1-Q3])	P
Sex(M/F)	–	2/18	1/19	1.000
Age	–	24.150 ± 2.142	24.200 ± 1.694	0.935
OSDI	–	29.639 ± 9.820	31.311 ± 8.889	0.576
NIBUT(s)	OD	6.690 [3.633–14.295]	5.925 [4.060–7.363]	0.310
	OS	7.935 [4.685–11.423]	6.880 [4.108–9.133]	0.337
FTBUT(s)	OD	2.610 [2.455–3.773]	2.630 [2.165–3.365]	0.457
	OS	2.840 [2.558–3.488]	2.790 [2.357–3.467]	0.465
TMH (mm)	OD	0.295 [0.250–0.375]	0.290 [0.253–0.328]	0.664
	OS	0.255 [0.230–0.300]	0.270 [0.260–0.305]	0.342
BRs(times/5 mins)	OD	106.000 [77.800–126.700]	113.500 [62.500–144.500]	0.957
	OS	105.500 [83.300–121.500]	109.500[61.500–138.500]	0.882
Percentage of IBs	OD	0.412 ± 0.231	0.475 ± 0.198	0.364
	OS	0.401 ± 0.232	0.494 ± 0.180	0.168
OPI	OD	0.975 [0.652–1.590]	0.815 [0.515–1.650]	0.441
	OS	0.950 [0.820–1.445]	0.740 [0.505–1.622]	0.239

Sex was compared using the chi-square test.

Age was assessed using an independent-samples t-test.

Continuous variables are presented as Mean ± SD if normally distributed, or as Median [Q1–Q3] if not.

All other baseline parameters were analyzed using Generalized Estimating Equations (GEE), with eye (OD/OS) specified as a within-subject factor to account for inter-eye correlation. Pairwise comparisons between groups were performed separately for right and left eyes using GEE-based multiple comparisons, with Bonferroni adjustment applied to control for type I error. This approach also accounted for potential interactions among time, group, and laterality, ensuring robust baseline comparisons.

*P < 0.05 indicates statistical significance.

Variation in OSDI scores over time

Within the experimental cohort, OSDI scores exhibited a significant time effect (P < 0.001). Day 30 values were significantly lower than baseline values from day 0 (P < 0.001). Although scores rose slightly by Day 60, the difference in OSDI scores between Day 60 and Day 0 remained statistically significant (P < 0.001) (Table 2).

Table 2.

GEE analysis of outcome measures across time in the experimental group

Parameters	Eye	Baseline	Day 30	Day 60	P(Wald χ²)	P1	P2	P3
OSDI	–	31.311 ± 1.937	19.247 ± 2.505	21.311 ± 2.894	<0.001*	<0.001*	<0.001*	0.815
NIBUT(s)	OD	6.347 ± 0.715	10.998 ± 1.614	9.469 ± 1.304	0.007*	0.019*	0.012*	0.803
	OS	8.156 ± 1.255	9.898 ± 1.476	7.715 ± 1.520	0.551	0.958	1.000	0.951
FTBUT(s)	OD	2.778 ± 0.159	3.229 ± 0.183	3.145 ± 0.171	0.014*	0.019*	0.176	1.000
	OS	2.833 ± 0.167	3.756 ± 0.217	3.294 ± 0.190	<0.001*	<0.001*	0.072	0.284
TMH (mm)	OD	0.293 ± 0.011	0.303 ± 0.013	0.299 ± 0.014	0.726	1.000	1.000	1.000
	OS	0.280 ± 0.009	0.296 ± 0.013	0.292 ± 0.015	0.535	0.903	1.000	1.000
BRs(times/5 mins)	OD	110.550 ± 13.896	133.850 ± 12.946	116.350 ± 13.029	0.043*	0.040*	1.000	0.642
	OS	108.950 ± 13.738	132.750 ± 12.492	116.800 ± 12.988	0.036*	0.034*	1.000	0.701
Percentage of IBs	OD	0.475 ± 0.043	0.239 ± 0.042	0.380 ± 0.037	<0.001*	<0.001*	0.061	0.002*
	OS	0.494 ± 0.039	0.226 ± 0.038	0.401 ± 0.036	<0.001*	<0.001*	0.043*	<0.001*
OPI	OD	0.999 ± 0.129	1.447 ± 0.170	1.191 ± 0.140	<0.001*	<0.001*	0.332	0.091
	OS	1.017 ± 0.144	1.684 ± 0.199	1.241 ± 0.143	<0.001*	<0.001*	0.300	0.020*

Data are presented as estimated marginal mean ± standard error (Mean ± SEM).

The Wald χ² associated P-values indicate the overall time effect in the GEE model, testing whether significant changes occurred over time across all three time points.

P₁ = comparison between Day 0 and Day 30; P₂ = comparison between Day 0 and Day 60; P₃ = comparison between Day 30 and Day 60.

P-values represent Bonferroni-adjusted pairwise comparisons among different time points.

*P < 0.05 indicates statistical significance.

Variation in NIBUT, FTBUT, and TMH over time

In the experimental group, a significant time effect was observed for Non-invasive tear breakup time (NIBUT) in the right eye (OD) (P = 0.007). NIBUT significantly increased from baseline (6.347 ± 0.715) to Day 30 (10.998 ± 1.614; P = 0.019). Although a slight decline was noted at Day 60, the values remained higher than the baseline. The left eyes (OS) showed no significant change.

Fluorescein tear breakup time (FTBUT) exhibited a significant time effect in both eyes (OD: P = 0.014; OS: P < 0.001). FTBUT improved significantly from baseline to Day 30 (OD: 2.778 ± 0.159 to 3.229 ± 0.183, P = 0.019; OS: 2.833 ± 0.167 to 3.756 ± 0.217, P < 0.001). By Day 60, these values had fallen and were no longer significantly different from the baseline.

Tear meniscus height (TMH) showed no significant time‑dependent changes in either eye (Table 2).

Variation in BRs and percentage of IBs over time

Generalized Estimating Equation (GEE) analysis revealed significant time effects for both BRs (OD: P = 0.043, OS: P = 0.036) and the percentage of IBs (both P < 0.001) within the experimental group. Compared to the baseline, BRs increased significantly by Day 30 (OD: P = 0.040, OS: P = 0.034) but partially regressed by Day 60. The percentage of IB fell markedly from baseline to Day 30 (both P < 0.001), and in OS, it remained lower at Day 60 (Table 2).

Variation in OPI over time

GEE analysis showed a significant time effect on the Ocular Protection Index (OPI) in both eyes (P < 0.001). The OPI increased significantly from baseline to Day 30 in both eyes (OD: 0.999 ± 0.129 to 1.447 ± 0.170, P < 0.001; OS: 1.017 ± 0.144 to 1.684 ± 0.199, P < 0.001). Although a reduction was observed at Day 60 (OD: 1.191 ± 0.140; OS: 1.241 ± 0.143), OPI values remained elevated compared to baseline. The difference between Day 0 and Day 60 was not statistically significant (OD: P = 0.332; OS: P = 0.300) (Table 2).

No significant temporal changes were observed in any parameter within the control group over the 30-day observation period (Table 3).

Table 3.

GEE analysis of outcome measures across time in the control group

Parameters	Eye	Baseline	Day 30	P
OSDI	–	29.639 ± 2.140	29.262 ± 3.100	0.863
NIBUT(s)	OD	8.442 ± 1.165	11.314 ± 1.555	0.087
	OS	9.752 ± 1.430	8.159 ± 1.310	0.360
FTBUT(s)	OD	3.104 ± 0.246	2.958 ± 0.140	0.533
	OS	3.151 ± 0.197	3.053 ± 0.176	0.634
TMH (mm)	OD	0.313 ± 0.020	0.303 ± 0.016	0.605
	OS	0.283 ± 0.018	0.292 ± 0.017	0.542
BRs(times/5 mins)	OD	116.850 ± 14.418	109.050 ± 12.409	0.510
	OS	115.950 ± 14.130	107.100 ± 11.855	0.429
Percentage of IBs	OD	0.412 ± 0.050	0.424 ± 0.051	0.776
	OS	0.402 ± 0.051	0.464 ± 0.054	0.074
OPI	OD	1.342 ± 0.283	1.085 ± 0.140	0.318
	OS	1.286 ± 0.239	1.103 ± 0.162	0.422

Data are presented as estimated marginal mean ± standard error (Mean ± SEM).

P-values represent Bonferroni-adjusted pairwise comparisons among different time points.

*P < 0.05 indicates statistical significance.

Correlation analysis

To explore potential predictors of subjective symptom changes, a GEE model was fitted, including sex, eyes (OD/OS), FTBUT, NIBUT, BRs, percentage of IBs, OPI, as well as key two-way interaction effects to account for potential interdependencies (Table 4). Sex showed a statistically significant main effect (β = –7.662, 95% CI: –13.236 to –2.088, P = 0.007), indicating lower OSDI scores in males compared to females. This result should be interpreted with caution due to the limited male sample size.

Table 4.

GEE model identifying predictors of OSDI score changes over time in the experimental group

Independent variables	β	95% CI	Wald χ2	P
	2.154	(−26.477, 30.785)	0.022	0.883
[Time=D30]	15.586	(0.910, 30.261)	4.333	0.037*
[Time=D0]	0a	.	.	.
Eye	−0.395	(−2.342, 1.551)	0.158	0.691
Sex	−7.662	(−13.236, −2.088)	7.259	0.007*
NIBUT(s)	−0.451	(−0.934, 0.032)	3.349	0.067
FTBUT(s)	6.992	(−2,177, 16.161)	2.234	0.135
BRs(times/5 mins)	0.238	(0.003, 0.472)	3.941	0.047*
Percentage of IBs	−7.246	(−27.771, 13.280)	0.479	0.489
OPI	−20.453	(−37.628, −3.277)	5.447	0.020*

95%CI = 95% confidence interval.

Values represent parameter estimates (β), 95% confidence intervals (CI), Wald χ² statistics, and corresponding P-values from a GEE model evaluating predictors of OSDI score changes across the intervention period. Time was modeled as a categorical variable (baseline = reference).

*P < 0.05 indicates statistical significance.

0_a: parameter considered redundant and set to zero.

OPI was significantly and inversely associated with OSDI (β = –20.453, 95% CI: –37.628 to –3.277, P = 0.020), indicating that greater ocular surface protection corresponded to reduced symptom severity. Conversely, BRs showed a modest but significant positive association with OSDI (β = 0.238, 95% CI: 0.003 to 0.473, P = 0.047), suggesting that higher BRs may not necessarily equate to symptom relief. No significant interaction was observed in this section.

Between-group comparisons at day 30

To evaluate the effect of the blink training intervention, GEE was employed to compare the experimental and control groups at Day 30 (Table 5). The interaction term group × time was statistically significant for several parameters, including OSDI (Wald χ² = 15.898, P < 0.001), FTBUT (Wald χ² = 9.107, P = 0.003), BRs(Wald χ² = 4.618, P = 0.032), percentage of IBs (Wald χ² = 29.627, P < 0.001), and OPI (Wald χ² = 8.542, P = 0.003).

Table 5.

GEE results for group-by-time interactions and eye-specific between-group comparisons at Day 30

Parameters	Main effect	2-way interaction (Time × Group)		3-way interaction (Time × Eye × Group)		Pairwise comparison at Day 30 (Group 1 vs. Group 2)
	Wald χ2, P	Wald χ2	P	Wald χ2	P	EYE	Wald χ2	P
OSDI	Time (18.013, <0.001)	15.898	<0.001*	–	–	–	6.649	0.010*
NIBUT(s)	Time (4.648, 0.031)	2.065	0.150	6.461	0.091	OD	0.006	0.938
						OS	0.921	0.337
FTBUT(s)	Time (4.443, 0.035), Eye (5.422, 0.020)	9.107	0.003*	10.173	0.017*	OD	1.285	0.257
						OS	6.228	0.013*
TMH (mm)	Eye (4.719, 0.030)	0.375	0.540	4.705	0.195	OD	0.000	0.982
						OS	0.042	0.838
BRs(times/5 mins)	Eye (8.458, 0.004)	4.618	0.032*	0.983	0.805	OD	2.058	0.151
						OS	2.381	0.123
Percentage of IBs	Time (16.321, <0.001) Sex (10.234, 0.001)	29.627	<0.001*	19.272	<0.001*	OD	7.543	0.006*
						OS	12.604	<0.001*
OPI	-	8.542	0.003*	17.888	<0.001*	OD	2.589	0.108
						OS	5.004	0.025*

The main effect columns indicate whether the effect of a single variable was statistically significant.

Two-way interactions (time × group) tested whether there were differences between the experimental and control groups over time.

Three-way interactions (time × eye × group) assessed whether the differences between the groups differed by eye (OD vs. OS).

Pairwise comparisons at Day 30 reflected whether there were significant differences between the experimental group (stratified by eye) and the control group at the main time points after the intervention.

*P < 0.05 was considered statistically significant and is marked with an asterisk.

Pairwise comparisons at Day 30 showed greater improvement in the experimental group for OSDI (P = 0.010), FTBUT (OS only, P = 0.013), percentage of IBs (OD: P = 0.006; OS: P < 0.001), and OPI (OS: P = 0.025). No between-group differences were observed in NIBUT, TMH, or BR at this time point. Three-way interactions (Time × Eye × Group) were significant for FTBUT, OPI, and IBs, suggesting that the intervention effects varied across eyes and over time.

To further evaluate inter-eye symmetry across time points, we included Supplementary Table 1, which showed pairwise comparisons between OD and OS within each group at D0 and D30. In most cases, no statistically significant differences were observed (Supplementary Table 1).

Discussion

The evolution of an information-driven society has led to a myriad of ocular discomfort among individuals frequently engaged in computer-related tasks. Prolonged VDT use is a salient risk factor for dry eye disease (DED)²⁰. A growing body of work has shown that blink interventions can relieve VDT-related DED symptoms^12,15,16. Building on these insights, the present study pioneered the application of smartphone software to modulate the blinking patterns of participants. This application operates in the background without significantly impeding work or leisure activities. In this study, young adults with DED were recruited, representing a population highly susceptible to screen-associated dry eye symptoms. After a month of smartphone-assisted training, individuals with DED experienced significant improvements: FTBUT was prolonged, OPI enhanced, BRs increased, and both OSDI scores and the frequency of IBs notably decreased. This regimen effectively minimized IBs. To assess durability, we re-evaluated the experimental group 30 days after stopping the software. Improvements in the OSDI score persisted, suggesting that the benefits endured beyond the training period. These findings indicate that smartphone-guided blink training can help mitigate DED symptoms and provide lasting clinical benefits.

Numerous empirical studies have elucidated that reduced blink efficiency results from a combination of lower total BR and a higher percentage of IB^21–23. The present study demonstrated a pronounced reduction in IB incidence. Such findings confirmed that diligent engagement with the software not only increased BR but also reduced IB prevalence, thereby enhancing overall blink efficiency, consistent with previous reports¹². The 2015 study by Nosch et al. introduced the Blink framework and primarily highlighted post‑training improvements in subjective symptoms; however, it lacked objective corroborative parameters¹⁴. In 2020, both Ashwini et al. and A.D. Kim et al. integrated objective ocular‑surface metrics, yet notable gaps remained: the former did not address potential changes in blinking patterns, and the latter did not definitively establish the efficacy of its training protocol in an exclusively DED cohort^12,15. In Kim’s study, participants frequently lapsed in adherence to the prescribed regimen, leading to inconsistent compliance and a higher attrition rate. By contrast, the robustness of the present study hinges on its innovative design, which minimizes reliance on participant self‑regulation. At each follow‑up visit, participants reported their average daily application usage, and investigators verified against a record booklet. Owing to the modern population’s dependence on smartphones, participants used the software well above the required daily duration. By harnessing software‑mediated reminders, participants received systematic prompts to perform blink training without intruding on daily activities. This streamlined approach, which demanded only a few hours of daily phone interaction, improved compliance and facilitated longitudinal monitoring.

One limitation of this study is the absence of an active control group, which restricts our ability to isolate the behavioral effects of blink training from potential placebo responses. While sham interventions, such as a non-functional app, would strengthen causal inference, designing a truly inert placebo condition is methodologically challenging. Because visual and attentional stimuli inherently influence blinking, even a minimally interactive interface could inadvertently affect blink behavior. Implementing a placebo that exerts no influence would likely require real-time synchronization with each participant’s natural blink rhythm, which is currently not technically feasible. Future studies should explore innovative solutions to address this limitation.

The observed improvements in OSDI, FTBUT, IBs, and OPI at Day 30 suggest that blink training may provide measurable short-term clinical benefits for individuals with DED. After 30 days of continuous training, the results showed significant improvements in most parameters in the experimental group compared to the baseline values, confirming the effectiveness of the software in improving the signs and symptoms of DED. First, the reduced OSDI scores indicated a decline in the subjective discomfort experienced by the participants. The significant reduction in OSDI scores also supported a perceived symptomatic benefit by the participants, even in the absence of full physiological normalization. Notably, although OSDI is commonly used to stratify symptom severity, with established thresholds (normal: 0-12; mild: 13-22; moderate: 23-32; severe: 33-100), our cohort exhibited minimal variability in baseline severity, with participants presenting moderate-to-severe scores (Table 1)²⁴. This lack of distributional spread limited the practical utility of formal subgroup classification in the current analysis.

Second, the improvements in FTBUT and OPI demonstrate the positive impact of blink training on ocular surface health, indicating enhanced tear film stability. Despite statistically significant increases in FTBUT, the post-intervention mean remained below the clinical threshold for normal tear stability (<5 s)^5,25. Given that FTBUT values ≥ 2 s and <5 s indicate mild to moderate DED, and <2 s suggests severe DED (particularly with corneal staining)²⁶. This indicates that the changes, while measurable, may not yet constitute a clinically significant recovery. Nevertheless, the observed improvements reflect a favorable trajectory toward improved tear dynamics, reinforcing blink training’s role as a behavioral adjunct rather than a curative intervention.

These improvements align with recent recommendations from the TFOS DEWS III and the Asia Dry Eye Society (ADES), which prioritize tear film stability as the primary therapeutic target and de-emphasize rigid severity classification due to frequent sign-symptom discordance^19,27. Consequently, we focused on group-level trends and symptomatic improvement, which hold greater clinical relevance for this demographically homogeneous, screen-dependent young adult population.

From a clinical perspective, these findings supported the feasibility of behavioral interventions, such as blink training, as adjunctive tools in DED management, particularly for screen-heavy populations. That exactly supported our intervention was designed as a behavioral adjunctive strategy rather than a therapeutic treatment for DED. Given that DED is multifactorial and chronic, full restoration of tear-film stability to normal clinical levels would not be expected after only a one-month blink-training programme. Moreover, the increase in blink frequency combined with the decrease in IBs emphasized the effectiveness of our training in modulating these participants’ blinking behavior. In future research, blink training may be considered as a complementary approach alongside standard clinical therapies, potentially enhancing overall treatment outcomes for DED.

To identify the key factors associated with symptom improvement during blink training, we applied a GEE model using the OSDI score as the dependent variable. The results revealed several significant predictors. Notably, the intervention time point at Day 30 was associated with a statistically significant improvement in OSDI (β = 15.586, P = 0.037), reinforcing the short-term effectiveness of the training protocol. Sex was also found to be a significant factor (β = –7.662, P = 0.007), with male participants reporting lower OSDI scores. However, given the extremely small number of male participants (n = 3), these findings must be interpreted with considerable caution. A notable limitation of this study is the pronounced sex imbalance in the sample, with the vast majority of participants being female. Although recruitment was open to all eligible participants without sex stratification, enrollment closure upon reaching the target sample size resulted in the unintended consequence of male underrepresentation. Consequently, the generalizability of the findings to males may be limited. Notably, the interaction between sex and treatment was not statistically significant in any outcome model, suggesting that the observed intervention effects were consistent across sexes within this sample. Nevertheless, the pronounced sex imbalance (predominantly female sample) necessitates caution in extrapolating these results to males, despite the statistical adjustment for sex in the GEE model. This recruitment imbalance likely reflects the well-documented phenomenon wherein females exhibit both a higher prevalence of dry eye disease and a greater willingness to participate in related clinical research^28–30. Therefore, to improve generalizability, future studies should ensure balanced sex representation through sex-stratified or quota-based recruitment.

OPI, defined by Eq. (1), showed a strong negative association with symptom severity (β = –20.453, P = 0.020), indicating that higher ocular surface protection—achieved via a more stable tear film relative to blink interval—was linked to greater symptom relief. This finding underscores the relevance of blink quality and tear stability in improving subjective comfort. Additionally, BR emerged as a positive predictor (β = 0.238, P = 0.047), suggesting that the increased blink frequency may reflect a compensatory response in symptomatic individuals. Together, these results highlighted that both behavioral parameters (such as BR) and ocular surface stability metrics (particularly OPI) might play essential roles in shaping DED symptomatology, providing mechanistic support for blink training as a non-invasive behavioral intervention.

The existing literature posits that a habit fundamentally manifests as an autonomic response to specific cues through consistent behavioral repetition in the presence of such stimuli^31–33. Nevertheless, health-behaviour interventions often exhibit short-lived efficacy. As individuals gradually deviate from the prescribed behavioral routine, the initial benefits diminish over time³⁴. In our study, after the 30-day intervention, participants were encouraged to forgo the software utilization for one month, after which a systematic data reassessment was undertaken. OSDI increased at Day 60 compared with Day 30; however, the score remained significantly lower than the baseline, consistent with expected habit decay. Parallel trends were noted in the objective measures: NIBUT in the OD and the percentage of IBs in OS demonstrated comparable partial reversals. Collectively, these findings suggested that while the blink training conferred sustained symptomatic relief relative to baseline, its efficacy in mitigating DED signs diminished over time post-intervention. In light of these insights, it can be postulated that extending the duration of the intervention may further consolidate blink efficacy^35,36. It is with regret that the current study was confined to a 30-day intervention period. Our team intends to prolong the duration of the intervention in subsequent investigations and devise diverse protocols tailored to the frequency of blink exercises. This approach aims to discern the optimal intervention strategy for the population with DED. Based on the findings of this study, extending the intervention could lead to significant improvements in key metrics, such as enhancing OPI and increasing FTBUT to more optimal durations, while the 30-day program lengthened FTBUT, confirming greater tear-film stability.

To provide a more nuanced evaluation of the therapeutic impact on the ocular surface in cases of DED, some academics have advocated for the OPI as a pivotal evaluative metric³⁷. In this study, we have incorporated OPI as an evaluative parameter, aiming to synergize blink interventions with clinical therapeutic modalities and to quantitatively assess the influence of blink training on tear film stability³⁷. The notable increase in the OPI after a 30-day blink training regimen, as evidenced in the experimental cohort, reaffirmed the strength of enhanced blink efficacy in bolstering tear film stability and safeguarding the ocular surface in patients with DED. One evident limitation of the present study is the omission of relevant meibomian gland parameters, rendering it challenging to pinpoint the source of the observed improvements in tear film stability.

Mobile healthcare and its digital therapeutic activities have garnered widespread attention from clinicians and healthcare providers due to its promising role in disease prevention, diagnosis, and treatment^38,39. In the burgeoning field of digital therapeutics, this study represents an innovative method that utilizes small-screen mobile phone-based software as a medium to improve blink efficacy via daily training modalities. Blink training has a distinct function, especially as a primary intervention for DED associated with VDT use or for those presenting with aberrant blinking metrics. The implementation of smartphone-based training software not only enables standardized, accessible training pathways but also, as this study confirms, significantly surpasses traditional blink training techniques in overall efficacy. Accumulating evidence has underscored the efficacy of smartphone-mediated interventions, positioning them as a strong complement to clinical regimens. In forthcoming investigations, our objective is to meticulously refine and quantify the ideal frequency and duration of blink training, to harmonize this intervention with clinical practices to achieve enhanced therapeutic outcomes. At the same time, the inclusion of DED subjects will be increased, and the population will be classified according to the subtype. Future studies will embark on stratified research targeting distinct classifications of DED, establishing tailored blinking frequencies, and incorporating comprehensive quantitative assessments of meibomian gland function and lipid layer integrity. Through the integration of this software’s design with conventional clinical therapies, we aspire to assess its efficacy and ascertain its value within clinical settings, thereby optimizing patient-specific treatment strategies for DED conditions.

Methods

Subject recruitment

The participants were thoroughly briefed on the study objectives and associated risks. Written informed consent was obtained from all the eligible participants. Participants were voluntarily recruited from Wenzhou Medical University through poster advertisements and class announcements. All participants had similar daily visual demands, environmental exposures, and dietary habits.

A total of 40 participants (80 eyes) were enrolled and randomly assigned to either the experimental group (n = 20) or the control group (n = 20), consisting of 3 males and 37 females. Randomization was conducted using a simple lottery method, in which participants drew one of 40 folded paper slips (20 per group) to determine allocation. The participants were all university students with a mean age of 24.1 ± 1.8 years (range: 18–30 years), and there was no significant difference between groups in age and sex ratio (P > 0.05).

Inclusion and Exclusion Criteria

Inclusion criteria were as follows: 1) age ≥ 18 years; 2) mobile phone use of ≥ 4 hours daily for ≥ 5 days weekly; 3) diagnosis of DED confirmed by FTBUT < 5 s and OSDI ≥ 13^3,27.

The exclusion criteria were as follows: 1) patients with a history of ocular diseases or surgeries; 2) patients with systemic diseases that might influence the study outcomes; 3) patients with prolonged use of sleeping pills, sedatives, or contraceptives, given their potential to affect blink efficiency and confound the results.

The application: WinkWell (Version 1.0, in-house development)

A novel smartphone application, “WinkWell”, was developed specifically for this study. The application features a black bar set at 10% of the screen width, with 25% opacity. This bar periodically appears every 5 s and remains visible for 1.5 s, prompting the user to blink twice. This design configuration was used to make the animation noticeable to users without hindering their ongoing tasks or concentration (Fig. 1).

Fig. 1. A schematic diagram of the electronic end of the application—WinkWell (version 1.0, self-research and development).

Each animated bar occupied 10% of the mobile phone screen, giving a total coverage of 20%. The bars were rendered with an opacity of 25%. The animation was triggered every 5 s and lasted 1.5 s. This configuration was designed to attract users’ attention without distracting them from their primary task. Moreover, the program required only minimal working-memory resources.

Blinking exercises

Blinking exercises were based on previously published recommendations¹⁴. Upon the appearance of the black bar, the participants were instructed to blink twice, ideally with a forceful squeeze¹². Those in the experimental group were directed to use the “WinkWell” application for ≥4 hours daily and ≥5 days weekly. The participants were instructed to discontinue exercise after completing 30 days of regular training. On the 60th day (30 days after the end of training), to explore the continuous effect of the training, we conducted a supplemental experiment with the participants in the experimental group.

Examination and image analysis

All ocular assessments were performed in a room where the temperature was maintained between 20 °C and 25 °C, and the humidity ranged from 40% to 60%. The participants underwent a series of examinations in the following sequence: (1) OSDI questionnaire, (2) NIBUT, (3) TMH, (4) FTBUT, and (5) blink pattern video recording (for BRs and percentage of IBs).

A trained examiner, who also implemented the random group allocation using a folded-paper-slip lottery method, conducted all clinical assessments, including the OSDI questionnaire, FTBUT, and ocular-surface imaging. Image analyses were performed independently by a second researcher who was also blinded to group assignments and clinical outcomes. Additionally, blink videos were reviewed by two other independent, trained raters who were blinded to both the intervention group and the study time points. They manually quantified BRs and the percentage of IBs. In total, four examiners were involved in data collection and analysis, each operating under blinded conditions to minimize assessment bias.

The OSDI questionnaire assessed ocular surface symptoms, with a score of ≥13 indicating clinical DED-related symptoms^27,40. TMH, and NIBUT were measured with the Keratograph 5 M (K5M, OCULUS, Inc., Wentzler, Germany)⁴¹.

The OPI quantifies the interplay between the interblink interval (IBI) and FTBUT and serves as a tool for assessing factors that influence tear-film instability associated with DED. The formula for the OPI is as follows (Eq. (1)):

$$\mathrm{Ocular\; Protection\; Index}\left(\mathrm{OPI}\right)=\frac{\mathrm{Tear\; Film\; Break}-\mathrm{up\; Time}\left(\mathrm{TBUT}\right)}{\mathrm{Inter\; Blink\; Interval}\left(\mathrm{IBI}\right)}$$ 1

An OPI of <1.0 signifies that the patient’s ocular surface is exposed, predisposing the eye to aggravated DED signs and symptoms. Conversely, an OPI of ≥1.0 indicates tear film protects the ocular surface and is typically associated with milder signs and symptoms. The OPI has been validated as a measure for tracking the evolution of DED severity over time and for gauging the efficacy of treatments aimed at enhancing tear film stability³⁷. In this study, OPI was used to quantify the therapeutic impact of blink training on the ocular surface.

The primary outcome measures were the parameters relevant for DED diagnosis, namely, OSDI and FTBUT. The percentages of IBs and the OPI served as secondary outcome measures.

Video recording

Video recordings were captured using a tablet (BTV-W09; Huawei Technologies Co., Ltd., China) positioned carefully behind a computer monitor. Participants were instructed to watch a 15-minute film segment presented on a Lenovo XiaoXin Air14IML laptop (Lenovo Beijing Co., Ltd., China). To simulate everyday conditions, participants were asked to sit comfortably and face the screen during recording (Fig. 2). All other electronic devices (e.g., mobile phones) were switched off to prevent external distractions. The segment was deliberately chosen to maintain a neutral affective domain, thereby mitigating potential emotional variables that might influence the blink dynamics of the participants. Given the plausible bias introduced by the participants’ cognizance of being observed, the initial 7 minutes of footage were excluded from analytical consideration⁴². A randomly selected 5-minute interval—considered optimal for blink-pattern analysis—was then extracted for evaluation⁴³.

Fig. 2. Positioning of the Recording Device.

Participants were seated comfortably at a desk with the computer positioned approximately 50 cm from their eyes. A tablet used for screen recording was positioned behind the computer monitor. (Written informed consent was obtained from the participant shown in the photograph for its publication.).

Blink dynamics were analyzed using video-editing software Adobe Premiere Pro-2020 (Adobe Systems, Inc., USA), which allowed the footage to be slowed for accurate counting. A full blink was recorded upon a confluence of the upper and lower eyelids. Conversely, an IB was recorded in the absence of complete closure of the upper eyelid⁸. Blink counts were performed independently by two trained researchers, and the mean of their results was used. The total number of blinks during the 5-minute clip was expressed as BR, and the number of IBs was expressed as the percentage of total blinks.

Ethics approval and consent to participate

Written informed consent was obtained from all participants prior to their inclusion in the study.

This prospective follow-up study was approved by the Human Participants Ethics Committee of Wenzhou Medical University (Reference:2021-222-K-195-01). All investigations adhered to the principles of the Declaration of Helsinki. Participants were thoroughly briefed on the study’s objectives and associated risks. Written informed consent was obtained from all eligible participants.

Statistical analysis

Statistical analyses were conducted using the Statistical Package for Social Sciences (SPSS) for Windows (version 21.0, IBM Corp., Armonk, NY, USA).

Sample size determination was performed with PASS software (version 15.0; NCSS Corp., Kaysville, UT, USA) based on two independent sample means with parameters α = 0.05 and power = 0.9. To detect clinically meaningful differences in BRs between groups, at least 16 eyes per group were required.

Continuous variables were assessed for normality using the Shapiro–Wilk test. Data conforming to a normal distribution are reported as mean ± standard deviation (SD), and age was compared between groups using the Independent-samples t-test. The chi-square test assessed sex distribution disparities. Non-normally distributed data are presented as Median (Q1-Q3).

GEE was applied to analyze group and time effects for all major outcomes, while accounting for within-subject correlations (inter-eye data), covariates (age and sex), and the study’s unbalanced repeated measures design. All inferential assessments adopted a two-tailed approach with a P-value threshold of < 0.05, considered statistically significant.

Author contributions

Z.X. and J.S.: Conceptualization, Methodology, Software. Z.X., J.S., and M.J.: Data curation, Writing- Original draft preparation. M.J, Y.W. and Y.S.: Visualization, Investigation. L.H.: Supervision. J.S. and Z.X.: Software, Validation. J.S., Z.X., F. L .and L.H.: Writing- Reviewing and Editing. All authors have read and approved the final manuscript.

Data availability

The datasets and the underlying code for this study were generated and/or analyzed during the current study and are not publicly available for proprietary reasons. Still, they are available from the corresponding author upon reasonable request.

Code availability

The custom code/scripts used for data generation and analysis in this study are not publicly available at this time. However, they can be obtained from the corresponding author upon reasonable request. The software used for the blink training application is WinkWell (Version 1.0, In-house Development).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-025-02053-8.