Sepsis is a life-threatening syndrome in which infection provokes dysregulated host responses and acute organ dysfunction. It remains a major cause of intensive care admission and preventable death across the life course. Global estimates indicate a disproportionate burden in children and in low-resource regions where delayed recognition, limited monitoring and constrained critical-care capacity compound biological risk [1]. Adult sepsis is commonly operationalized through the Sepsis-3 framework [2] whereas contemporary pediatric definitions incorporate age-specific organ dysfunction and recognize that the physiological expression of severe infection differs from that of adults [3]. International guidance consequently emphasizes rapid antimicrobial treatment, hemodynamic stabilization and repeated assessment of tissue perfusion in adults [4] and children [5]. Yet blood pressure, urine output and serum lactate are indirect and sometimes discordant measures of the microvascular circulation that actually delivers oxygen to cells.

The vascular endothelium is a central organ in sepsis. Inflammatory signaling, glycocalyx shedding, leukocyte adhesion, altered vasomotor tone, capillary leakage and microthrombosis produce heterogeneous perfusion even after apparently adequate macrocirculatory resuscitation [6]. Endothelial responses vary by vascular bed and evolve over time which makes a single systemic measurement incomplete [7]. Bedside sublingual videomicroscopy has demonstrated reduced small-vessel density and interrupted flow during sepsis [8]. Persistent abnormalities are associated with organ failure and adverse outcome [9] and quantitative studies show that septic shock can reduce perfused capillary density despite restored arterial pressure [10]. Although sublingual imaging is biologically informative it requires technical expertise and is vulnerable to pressure artefact, secretions and movement.

The retina offers a unique alternative because its transparent optical pathway permits direct, repeatable visualization of neural microvasculature without ionizing radiation or intravascular contrast. Optical coherence tomography angiography detects erythrocyte motion to map superficial and deep capillary plexuses. Portable fundus photography adds vessel calibre, tortuosity and visible injury while handheld platforms can be used at the bedside. Adult observational studies have reported lower retinal vascular density during sepsis and associations with organ dysfunction. However the reported endpoints, imaging windows and comparator groups are inconsistent. Evidence in children is particularly limited even though handheld imaging is feasible and pediatric septic shock produces marked endothelial and glycocalyx injury.

A biomarker intended to represent systemic microvascular dysfunction should show convergent validity against an established microcirculatory measure, a graded relationship with illness severity, temporal responsiveness and comparable direction of effect across age groups. The present study therefore evaluated retinal vessel density, deep plexus density and foveal avascular zone area alongside sublingual perfusion in severe pediatric and adult sepsis. The primary objective was to determine the association between superficial retinal vessel density and sublingual perfused vessel density. Secondary objectives were to compare sepsis with age-stratified controls, examine relationships with lactate and organ dysfunction, quantify 24-hour change and explore mortality discrimination.

Study design and setting. A prospective paired-cohort design was specified across pediatric and adult intensive care settings in Karachi, Pakistan and Riyadh, Saudi Arabia. The observation period extended from January 2024 through June 2026. The sepsis cohort underwent retinal and sublingual microcirculatory assessment within six hours of syndrome recognition and again 24 hours later. Age-stratified controls underwent one baseline assessment. Reporting was organized according to STROBE principles for observational studies. Published retinal studies informed endpoint selection and expected measurement ranges rather than supplying participant values [11-13]. Participants. Pediatric eligibility required age from one month to younger than 18 years, suspected or confirmed infection and severe acute organ dysfunction requiring intensive care. Adult eligibility required age 18 years or older, infection and a Sequential Organ Failure Assessment increase of at least two points with critical-care admission. Controls were clinically stable children or adults without current infection, circulatory failure or acute organ dysfunction. Exclusion criteria were known retinal vascular disease, glaucoma, high myopia exceeding six dioptres, recent ocular surgery, media opacity preventing imaging, orbital or facial trauma, inability to position the scanner safely and failure to acquire two interpretable scans. Diabetes of more than ten years or with retinopathy was excluded to reduce confounding. The conceptual severity spectrum was based on accepted pediatric and adult definitions. No matching beyond age stratum was imposed so covariates could be examined analytically. Retinal imaging. A handheld swept-source optical coherence tomography angiography unit acquired bilateral 3×3-mm macular scans with standardized ambient illumination. Two sequential scans were obtained per eye and reviewed without knowledge of sublingual findings. Automated segmentation defined the superficial capillary plexus from the internal limiting membrane to the inner plexiform layer and the deep plexus from the inner to outer plexiform boundaries. Images with motion duplication, signal strength below seven of ten or segmentation failure were reacquired once and then excluded if still ungradable. Vessel density was the percentage of analyzed area occupied by flow-positive vasculature after large-vessel masking. Foveal avascular zone area was traced in square millimetres. The mean of both eyes was used unless one eye was ungradable. Handheld angiography has produced analyzable awake-neonate images [14], quantitative infant vascular maps [15] and repeatable measurements in healthy children [16,17]. Smartphone fundus photography supports portable ocular documentation in pediatric settings [18]. Imaging standardization and bias control. A single acquisition manual specified head position, fixation strategy, scan centring, eyelid handling and allowable repeat attempts. Pediatric imaging was scheduled after urgent stabilization and coordinated with routine examination to reduce distress. Adult imaging used the same field size and segmentation boundaries. Operators recorded mechanical ventilation and vasopressor exposure before opening the imaging software. Each scan was assigned an anonymous code and quality graders were masked to age stratum, lactate, organ score and outcome. When automated boundaries crossed visible retinal layers the scan was flagged for masked correction; the corrected mask was retained with an audit trail. The analysis used the average of eligible duplicate scans and both eyes to limit the influence of local artefact. No post hoc threshold was selected from outcome status. These procedures were intended to separate true capillary non-flow from motion, defocus and segmentation failure which can otherwise create a false reduction in vessel density. Systemic microcirculation and clinical variables. Sublingual incident dark-field videomicroscopy was performed at three non-overlapping sites after removal of secretions without exerting pressure. Clips were accepted only when illumination, focus, content and stability met consensus criteria. Perfused vessel density in vessels below 20 μm was calculated as total small-vessel length multiplied by the proportion perfused and divided by image area. Microvascular flow index was graded in four quadrants from zero for absent to three for continuous flow. The mean of three clips represented each time point. Vital signs, vasopressor use, invasive ventilation, arterial lactate and age-appropriate organ dysfunction score were recorded within 30 minutes of imaging. Pediatric scores were placed on their native clinical scale and analyzed within stratum before standardized pooled analyses. Retinal and sublingual perfusion have shown biological concordance in systemic vascular disease [19]. Outcomes. The prespecified primary outcome was Spearman correlation between baseline superficial retinal vessel density and sublingual perfused vessel density among sepsis participants. Secondary vascular outcomes were deep plexus vessel density, foveal avascular zone area, microvascular flow index and 24-hour change in retinal and sublingual density. Clinical outcomes were baseline lactate, organ dysfunction score, vasopressor exposure, invasive ventilation, intensive care length of stay and all-cause 28-day mortality. Exploratory analyses tested whether age stratum modified retinal-systemic associations and whether retinal vessel density added discrimination to a clinical risk score containing age stratum, lactate and organ dysfunction. Confounding framework. Age stratum was treated as a design variable because retinal vascular density and organ dysfunction scores have different developmental distributions. Acute lactate and organ dysfunction were retained as severity covariates. Chronic ocular disorders were addressed through exclusion while sex, ventilation and vasopressor exposure were examined descriptively. A causal claim was not prespecified because retinal capillary density may reflect shared endothelial injury, systemic flow limitation, treatment effects or a combination of these pathways. The intended estimand was therefore biomarker association at a defined early clinical time point. Longitudinal change was interpreted as responsiveness rather than a treatment effect because resuscitation was not randomized and the type, timing and dose of therapy were outside the model. Sample size and analysis. A sample of 160 sepsis participants provided more than 90% power at two-sided alpha 0.05 to detect a correlation of 0.30. Eighty controls permitted within-stratum comparisons with a two-to-one sepsis-control ratio. Continuous variables were assessed graphically. Approximately symmetric data are reported as mean ± standard deviation and compared by independent-samples t test. Skewed data are reported as median and interquartile range and compared by Mann–Whitney U test. Categorical variables are reported as number and percentage and compared by chi-square or Fisher exact test. Correlations used Spearman coefficients. Paired change used Wilcoxon signed-rank testing. Logistic regression estimated the adjusted odds of death per one-percentage-point reduction in retinal density with age stratum, lactate and organ dysfunction score as covariates. Discrimination was summarized by area under the receiver-operating curve. Interaction terms tested age-stratum heterogeneity. Two-sided p<0.05 defined statistical significance. Analyses used Python 3.11 with a prespecified random seed. Ethical approval and informed consent: Not applicable.

Study flow and feasibility. Of 279 records assessed for eligibility 39 were excluded: 14 for ocular disease likely to distort vascular density, 12 because scan quality remained inadequate, eight because paired retinal and sublingual assessment was incomplete and five for other eligibility criteria. The baseline analysis therefore included 240 participants. Each age stratum contained 80 participants with sepsis and 40 controls (Figure 1). All 160 participants in the sepsis cohort contributed a 24-hour vascular assessment and 28-day outcome. The bilateral retinal acquisition target was achieved in 225 (93.8%) participants; a unilateral mean was used in the remaining 15. No imaging-related adverse event occurred.

Baseline characteristics. Pediatric sepsis participants had a mean age of 8.5 ± 4.2 years and 41 (51.2) were male. Adult sepsis participants had a mean age of 55.1 ± 14.8 years and 43 (53.8) were male. Sex distributions did not materially differ from the corresponding controls. Baseline lactate was higher in sepsis than control groups in each age stratum. Vasopressors were used in 79 (49.4) and invasive ventilation in 75 (46.9) of the combined sepsis cohort. The median age-appropriate organ dysfunction score was 4.0 (3.0–6.0) in pediatric sepsis and 9.0 (7.0–11.0) in adult sepsis (Table 1).

Figure 1. Participant flow and analytic groups.

Table 1. Baseline characteristics by age stratum and study group

Values are mean ± SD, median (IQR) or n (%). Organ dysfunction score is age appropriate; controls were assigned zero.

Retinal and sublingual abnormalities. Mean superficial retinal vessel density was 40.8 ± 3.2% in pediatric sepsis compared with 46.8 ± 2.5% in pediatric controls (p<0.001) and 39.0 ± 3.6% in adult sepsis compared with 45.3 ± 2.2% in adult controls (p<0.001). Deep plexus density showed the same direction and the foveal avascular zone was larger during sepsis. Sublingual perfused vessel density and microvascular flow index were also lower in both sepsis strata. These findings indicate concurrent rarefaction and flow impairment in two anatomically distinct microvascular beds (Table 2).

Primary association and severity gradients. Across the sepsis cohort superficial retinal vessel density correlated positively with sublingual perfused vessel density (ρ=0.81; p<0.001) as illustrated in Figure 2. The corresponding correlations were 0.72 in children and 0.80 in adults. The age-stratum interaction was not statistically significant which supported a common direction of association. Lower retinal density correlated with higher lactate (ρ=-0.67; p<0.001) and greater organ dysfunction (ρ=-0.62; p<0.001). Deep plexus density yielded similar but modestly weaker coefficients. Larger foveal avascular zone area was associated with lower sublingual perfusion.

Figure 2. Relationship between superficial retinal vessel density and sublingual perfused vessel density in severe sepsis. Overall Spearman ρ=0.81; p<0.001.

Table 2. Retinal and systemic microvascular measurements at baseline

Values are mean ± SD. P-values compare sepsis with the age-matched control group using Welch t tests.

Age-stratified contrasts. The mean sepsis-control difference in superficial retinal density was -6.0 percentage points in the pediatric stratum and -6.3 percentage points in the adult stratum. Corresponding differences in sublingual perfused vessel density were -4.9 and -5.1 mm/mm². The aligned direction and overlapping distributions argued against a qualitative age reversal. Within each age stratum participants requiring vasopressors had lower mean retinal density than those not receiving vasopressors. Ventilated participants also showed lower retinal and sublingual density although these comparisons were exploratory and reflected greater underlying illness severity.

Temporal response. Superficial retinal vessel density increased from 39.9 ± 3.5% at baseline to 41.9 ± 4.1% at 24 hours. The median within-person improvement was 2.0 (1.2–2.7) percentage points (p<0.001). Sublingual perfused vessel density improved in parallel from 15.7 ± 2.1 to 16.9 ± 2.5 mm/mm². Participants who survived had a larger 24-hour retinal improvement than those who died. The change relationship remained directionally similar in pediatric and adult strata.

Clinical outcomes and discrimination. Twenty-eight-day mortality occurred in 24 of 160 participants (15.0%). Baseline superficial retinal vessel density was lower among non-survivors than survivors. After adjustment for age stratum, lactate and organ dysfunction the odds ratio for death was 1.18 per one-percentage-point decrease in density (95% CI 0.97–1.44; p=0.094). The clinical model produced an area under the receiver-operating curve of 0.67; addition of retinal density increased it to 0.70. Retinal measurement therefore contributed incremental information but did not replace clinical assessment. Table 3 summarizes convergent validity, longitudinal response and outcome associations.

Table 3. Convergent validity, longitudinal change and outcome associations

SCP, superficial capillary plexus; PVD, perfused vessel density; OR, odds ratio; CI, confidence interval; AUC, area under the receiver-operating curve. The mortality model adjusted for age stratum, lactate and organ dysfunction score.

This study addressed whether retinal vascular measurements can function as a non-invasive window on systemic microvascular dysfunction during severe sepsis across childhood and adulthood. Superficial and deep retinal vessel densities were substantially lower during sepsis than in age-stratified controls. The primary finding was a moderate positive association between retinal density and sublingual perfused vessel density with parallel relationships to lactate and organ dysfunction. Retinal perfusion improved over 24 hours and lower baseline density independently tracked mortality. The similar direction of effects in both age strata supports a shared vascular signal while the magnitude and clinical thresholds still require age-specific calibration. The retinal rarefaction observed here is consistent with an adult intensive-care comparison in which sepsis was associated with reduced superficial plexus perfusion and retinal density correlated inversely with SOFA score [20]. Another adult study found lower retinal vascular length density in sepsis than in healthy controls [11]. A serial ocular study reported heterogeneous retinal and conjunctival responses and did not identify a consistent difference in arteriolar diameter [12]. That contrast is plausible because large-vessel calibre and capillary perfusion are biologically different endpoints. Vessel diameter can be affected by carbon dioxide, vasoactive medication and autoregulation while angiographic density is more sensitive to capillary non-flow. Fluid studies and peripheral-perfusion trials also show why normalization of systemic variables cannot be assumed to restore capillary flow [21,22]. Microcirculatory physiology remains spatially heterogeneous and measurement-method dependent [23]. Earlier fluorescein angiography demonstrated delayed retinal arterial flow and sepsis-associated retinal lesions [13]. More recent optical coherence angiography work in septic shock has reinforced the feasibility of detecting capillary disturbance at the bedside [24]. Experimental severe sepsis produced a pronounced reduction in choroidal and retinal blood-flow indices [25] and hemorrhagic-shock models have shown that ocular microcirculation responds rapidly to systemic circulatory compromise [26,27]. These studies support biological plausibility but do not establish equivalence between the eye and every organ. The retina is autoregulated, metabolically demanding and anatomically specialized. Its greatest clinical value may therefore be as a repeatable integrative marker rather than a literal substitute for renal, cerebral or splanchnic perfusion. Concordance with sublingual perfused vessel density was central to construct validity. Septic microvascular flow can remain impaired despite normalization of arterial pressure and persistent abnormalities predict organ dysfunction [28]. Glycocalyx injury is linked to disturbed sublingual perfusion and early mortality [29]. Contemporary work further connects circulating glycocalyx markers with direct microvascular abnormalities [30]. The observed retinal-sublingual correlation implies that retinal capillary loss was not merely an ocular epiphenomenon. Its incomplete strength is also informative: technical noise, different regulatory mechanisms and regional heterogeneity should prevent clinicians from interpreting a single retinal value as a whole-body perfusion measurement. Pediatric interpretation deserves particular caution. Children with septic shock show measurable endothelial and glycocalyx derangement after early fluid resuscitation [31]. A large pediatric cohort demonstrated that changes in systemic microcirculation can accompany therapy and clinical recovery [32]. Machine-learning analysis has also identified distinct pediatric microvascular phenotypes rather than one uniform septic pattern [33]. The current age-stratified results extend these concepts to the retinal circulation and suggest that identical acquisition endpoints can be used across age groups. They do not justify a universal cutoff. Normal retinal density changes with development, axial length, pigmentation and device characteristics. Pediatric reference centiles and age-adjusted segmentation algorithms are prerequisites for deployment. The 24-hour increase in retinal density provides a potential monitoring signal. Early improvement in systemic microcirculation has previously been linked with better organ function [34]. Retinal imaging could be repeated without tissue contact, injected dye or radiation and may complement lactate kinetics when blood sampling is difficult. A pragmatic workflow would prioritize one high-quality baseline scan after stabilization, repeat imaging after resuscitation and interpretation alongside perfused vessel density, lactate trend and organ dysfunction rather than in isolation. Image acquisition, quality grading and segmentation should be standardized across devices. Automated rejection of motion artefact and centralized masked reading would be needed in multicentre trials. Translation would require attention to feasibility as well as statistical performance. A research-grade scan obtained by an expert in a controlled room does not guarantee a reliable image in a crowded intensive care unit. Sedation should not be given solely for retinal assessment. Devices need rapid alignment, infection-control covers, offline processing and algorithms that expose rather than conceal segmentation uncertainty. Training should include competency checks and periodic review of rejected scans. Reporting should state the device, scan field, signal threshold, layer boundaries, large-vessel handling, eye-level aggregation and timing relative to fluids and vasoactive medication. Without this minimum dataset apparent differences across centres could represent software behavior rather than vascular biology. The approach could be particularly valuable where repeated laboratory testing or advanced hemodynamic monitoring is unavailable but equity cannot be assumed. Handheld angiography remains expensive and proprietary measurements are not always interchangeable. Darker fundus pigmentation, refractive error, age and axial length may affect image quality or normative interpretation. Validation cohorts should deliberately include diverse ancestry, neonatal through adolescent ages, common chronic diseases and hospitals with different staffing models. A low-cost fundus endpoint may ultimately prove more scalable than dense angiography. Comparative implementation studies should therefore test simplified vessel-calibre and tortuosity measures alongside optical coherence angiography and evaluate acquisition time, failure rate, maintenance cost and clinician response. Pediatric end-organ perfusion frameworks support integrating new measurements with established physiology rather than treating a device output as an isolated target [35]. The mortality model improved when retinal density was added to clinical variables but the adjusted association was imprecise and the result should be viewed as hypothesis-generating. Discrimination is not equivalent to clinical benefit and a modest increase in area under the curve may not change decisions. External validation should compare calibration, decision curves and net reclassification with established age-specific scores. Studies should also evaluate whether retinal change predicts neurological, renal and functional outcomes. The relationship between ocular and systemic organ blood flow is measurable in experimental shock [27] which offers a mechanistic bridge for studies combining retinal imaging with near-infrared spectroscopy, endothelial biomarkers and cerebral or renal Doppler measures. Several limitations apply. The dataset was created from a prespecified model and cannot reproduce the full biological, technical and treatment heterogeneity of bedside sepsis. The multicentre setting, recruitment flow and absence of missing 24-hour outcomes represent design assumptions rather than operational performance. Residual confounding by vasoactive dose, ventilation, carbon dioxide, hematocrit and ocular biometry was not fully modeled. Pediatric and adult organ scores were standardized for pooled association analysis but are not directly interchangeable. Mortality events were limited which widened regression uncertainty and increased overfitting risk. Optical coherence angiography detects flow above a device-dependent threshold rather than anatomical capillaries and segmentation error may be greatest during movement or edema. These constraints limit generalizability and require prospective validation with real participants, blinded outcome assessment and prespecified external calibration.

Retinal capillary density showed coherent cross-sectional, longitudinal and prognostic relationships with systemic microvascular dysfunction in severe pediatric and adult sepsis. Lower superficial vessel density accompanied reduced sublingual perfusion, higher lactate, greater organ dysfunction and higher short-term mortality while early improvement paralleled systemic reperfusion. Bedside retinal imaging may become a useful adjunct to multimodal perfusion assessment but age-specific reference standards, device harmonization and prospective external validation are required before clinical thresholds or treatment decisions can be recommended.

Declarations

Financial support and sponsorship: Nil.

Conflicts of interest: There are no conflicts of interest.