This cohort study identifies a quantitative characteristic of leukocyte-endothelial interactions after hematopoietic cell transplantation and tests its associations with patient outcomes.
Key Points
Question
Are leukocyte-endothelial interactions in skin associated with patient outcomes after hematopoietic cell transplantation for hematologic cancer?
Findings
In this cohort study of 56 patients, leukocyte-endothelial interactions assessed noninvasively by confocal videomicroscopy had statistically significant associations with relapse, relapse-free survival, and overall survival. These interactions accounted for 82% to 95% of the prognostic information to predict each outcome; by contrast, the best existing clinical predictor routinely available accounted for 10% to 28% of the prognostic information in the same model.
Meaning
Assessing leukocyte-endothelial interactions could noninvasively and dynamically identify patients who are at high risk for severe outcomes of systemic disease.
Abstract
Importance
Hematopoietic cell transplantation (HCT) is a potential cure for hematologic cancer but is associated with a risk of relapse and death. Dynamic biomarkers to predict relapse and inform treatment decisions after HCT are a major unmet clinical need.
Objective
To identify a quantitative characteristic of leukocyte-endothelial interactions after HCT and test its associations with patient outcomes.
Design, Setting, and Participants
In this prospective single-center cohort study from June 2017 to January 2020, patients of any age, sex, race, and ethnicity who had HCT for hematologic cancer were referred by health care professionals as either suspected of having symptoms or not having symptoms of acute graft-vs-host disease between 25 and 161 days after HCT. Patients underwent noninvasive skin videomicroscopy. Videos of dermal microvascular flow were recorded with a reflectance confocal microscope. Two blinded observers (J.R.P. and Z.Z.) counted leukocytes adherent to and rolling along the vessel wall per hour (A&R). Of 57 enrolled patients, 1 relapsed before imaging and was excluded, resulting in 56 patients included in analyses.
Main Outcomes and Measures
Relapse of cancer, relapse-free survival, and overall survival.
Results
Among the 56 patients (median age, 59 years; 38 [68%] male) who underwent imaging a median of 40 days after HCT, 21 had high A&R and 35 had low A&R. After correcting for the revised Disease Risk Index, patients with high A&R had higher rates of relapse (hazard ratio [HR], 4.24; 95% CI, 1.32-13.58; P = .02), reduced relapse-free survival (HR, 3.29; 95% CI, 1.26-8.55; P = .02), and reduced overall survival (HR, 3.06, 95% CI, 1.02-9.19; P = .05). These associations were preserved after correcting for possible confounders, steroid treatment, and acute graft-vs-host disease status. In the prognostic adequacy calculation by using Cox models, the new imaging biomarker (A&R) accounted for 82% to 95% of the prognostic information to predict each outcome. By contrast, the best existing clinical predictor routinely available, the revised Disease Risk Index, accounted for 10% to 28% of the prognostic information in the same model.
Conclusions and Relevance
In this cohort study, leukocyte-endothelial interactions, visualized directly in skin after HCT, were associated with the patient outcomes of relapse, relapse-free survival, and overall survival. Assessing this dynamic marker could help patients at high risk for relapse who may benefit from interventions, such as early withdrawal of immunosuppression.
Introduction
Hematopoietic cell transplantation (HCT) leverages graft-vs-tumor effect, mediated by donor leukocytes to cure hematologic disease, and is the only potentially curative therapy for patients with poor-risk disease.1 Up to 70% of all deaths after HCT are associated with relapse.2 The revised Disease Risk Index (rDRI) has been associated with death due to relapse.3 Such traditional risk assessments are based on retrospective data and may be slow to reflect improved outcomes related to novel therapies. Furthermore, routinely available pre-HCT risk factors cannot be used as dynamic biomarkers for monitoring patient status. While markers exist for nonrelapse mortality (NRM)4 and overall survival (OS),5 dynamic and actionable biomarkers to predict relapse after HCT remain an unmet need.
A candidate biomarker for assessing patient risk after HCT is leukocyte-endothelial interactions. In humans, these interactions are associated with survival after septic shock6 and are higher in pathologic states.7 Two important steps in the leukocyte inflammatory cascade are leukocyte rolling and subsequent adhesion.8 Rolling is a low-affinity adhesive interaction, in which the force of blood flow induces a rotational motion. Adhesion, by contrast, is a high-affinity binding interaction, denoting the absence of movement.9 With emerging technologies, these leukocyte-endothelial interactions can be seen in the human eyes,7,10 sublingual mucosa,6 nailfold capillaries,11 and skin.12 However, methods to leverage these technologies and leukocyte-endothelial interactions for clinical benefit are not fully established.
The skin offers convenient access for monitoring leukocyte-endothelial interactions. Specifically, the upper dermal microvasculature can be visualized noninvasively by reflectance confocal videomicroscopy. Hundreds of published studies on tissue morphology show the use of reflectance confocal microscopy as a diagnostic “optical biopsy” for pigmented13 and inflammatory14 skin lesions. Although confocal microscopy can visualize individual blood cells moving in the microvasculature at video rate, to our knowledge, the clinical value of directly imaging leukocyte dynamics in humans via videomicroscopy remains unknown. More generally, the potential of using cutaneous imaging to prognostically evaluate systemic nondermatological disease remains underexplored.
Previously, we noninvasively acquired confocal microscopy images to describe histopathological features of acute graft-vs-host disease (aGVHD)15 after HCT.16 During the imaging, we also recorded videos of cell motion in the upper dermal microvasculature. Interestingly, we noticed that several patients with higher leukocyte-endothelial interactions died in the subsequent months, regardless of whether they had aGVHD.17 We hypothesized that leukocyte-endothelial interactions are associated with outcomes after HCT, and we aimed to compare the strength of this association and model adequacy with rDRI.
Methods
Study Participants
From June 2017 to January 2020, we prospectively recruited a cohort of patients of any age, sex, race, and ethnicity (Table 1) who had HCT for hematologic cancer at the joint program of Vanderbilt University Medical Center and the US Department of Veterans Affairs in Nashville, Tennessee. Patients were imaged as either suspected of having symptoms of aGVHD or as having no symptoms of aGVHD, as determined by a health care professional. The institutional review boards at Vanderbilt University and the Tennessee Valley Healthcare System granted approval of the study, and all patients gave written informed consent.
Table 1. Summary and Comparison of Characteristics Between Patients With High and Low A&R Counts.
| Characteristic | No. (%) | ||
|---|---|---|---|
| High A&R (n = 21) | Low A&R (n = 35) | All (n = 56) | |
| Age, median (IQR) [range], y | 61 (54-64) [28-71] | 58 (46-63) [16-72] | 59 (48-63) [16-72] |
| Sex | |||
| Female | 5 (24) | 13 (37) | 18 (32) |
| Male | 16 (76) | 22 (63) | 38 (68) |
| Disease | |||
| Acute lymphoblastic leukemia | 1 (5) | 4 (11) | 5 (9) |
| Acute myeloid leukemia | 10 (48) | 15 (43) | 25 (45) |
| Myelodysplastic syndrome | 6 (29) | 6 (17) | 12 (21) |
| Other | 4 (19) | 10 (29) | 14 (25) |
| HCT comorbidity index | |||
| 0 | 3 (14) | 10 (29) | 13 (23) |
| 1-2 | 10 (48) | 10 (29) | 20 (36) |
| ≥3 | 8 (38) | 15 (43) | 23 (41) |
| Donor match | |||
| Matched related donor | 8 (38) | 13 (37) | 21 (38) |
| Nonmatched related donor | 13 (62) | 22 (63) | 35 (63) |
| Conditioning | |||
| Myeloablative | 12 (57) | 16 (46) | 28 (50) |
| Reduced intensity | 9 (43) | 19 (54) | 28 (50) |
| Sex matching | |||
| Female to female | 2 (10) | 5 (14) | 7 (13) |
| Female to male | 2 (10) | 7 (20) | 9 (16) |
| Male to female | 3 (14) | 8 (23) | 11 (20) |
| Male to male | 14 (67) | 15 (43) | 29 (52) |
| Follow-up for survivors, median (IQR) [range], mo | 18 (10-23) [8-26] | 9 (9-20) [6-28] | 13 (9-20) [6-28] |
| aGVHD symptoms on imaging day | |||
| Yes, later definitively confirmed | 14 (67) | 18 (51) | 32 (57) |
| Yes, final diagnosis uncertain | 3 (14) | 1 (3) | 4 (7) |
| No | 4 (19) | 16 (46) | 20 (36) |
| Glucksberg grade | |||
| 0 (Confirmed absence of aGVHD) | 4 (19) | 16 (46) | 20 (36) |
| 1 | 8 (38) | 13 (37) | 21 (38) |
| 2 | 2 (10) | 2 (6) | 4 (7) |
| Unknown | 7 (33) | 4 (11) | 11 (20) |
| Receiving systemic steroids prior to and on imaging day | |||
| Yes | 7 (33) | 4 (11) | 11 (20) |
| No | 14 (67) | 31 (89) | 45 (80) |
| No. of days receiving steroids, median (IQR) [range] | 7 (3-30) [2-45] | 1 (1-5) [1-8] | 5 (2-14) [1-45] |
| No. of imaged body sites | |||
| 1 | 4 (19) | 5 (14) | 9 (16) |
| 2 | 16 (76) | 29 (83) | 45 (80) |
| 3 | 1 (5) | 1 (3) | 2 (4) |
| Imaged body sites | |||
| Left volar forearm | 15 (71) | 30 (86) | 45 (80) |
| Left upper anterior chest | 8 (38) | 24 (69) | 32 (57) |
| Revised Disease Risk Index | |||
| Very high | 2 (10) | 0 | 2 (4) |
| High | 9 (43) | 13 (37) | 22 (39) |
| Intermediate | 10 (48) | 20 (57) | 30 (54) |
| Low | 0 | 2 (6) | 2 (4) |
| A&R, median (IQR) [range] | 33 (20-50) [15-682] | 0 (0-6) [0-13] | 8 (0-22) [0-682] |
| Absolute neutrophil count, median (IQR) [range], ×1000/μL | 3.02 (2.50-4.84) [1.44-13.95] | 3.36 (1.89-3.97) [0.14-11.61] | 3.35 (2.31-4.28) [0.14-13.95] |
Abbreviations: aGVHD, acute graft-vs-host disease; A&R, total number of adherent and rolling leukocytes per hour; HCT, hematopoietic cell transplantation.
Imaging Protocol for Reflectance Confocal Videomicroscopy
We noninvasively visualized cutaneous microvessels with the VivaScope 1500 (Caliber I.D.), a reflectance confocal microscope approved by the US Food and Drug Administration (FDA).12 The microscope uses a near-infrared (830 nm), low-power (<20 mW) laser light to acquire 8 × 8-mm2 optical sections in tissue up to 0.2 mm deep at 0.7 μm lateral and 3 μm axial resolution. The microscope captures 0.5 × 0.5-mm2 fields of view at a video rate of 9 frames per second, enabling direct, real-time visualization of the upper dermal microvasculature.
Within an 8 × 8-mm2 confocal mosaic grid, we aimed to capture 30-second videos of 10 different 0.5 × 0.5-mm2 fields of view (ie, all corners and several central parts of the grid) at a depth of 60 to 100 μm. Some videos were shorter or longer owing to patient movement or interrupted imaging sessions, but overall imaging sessions followed the published protocol,18 regardless of aGVHD symptoms or other status during the imaging session. Per site, we collected 10 videos with visible blood flow or leukocyte-endothelial interactions, resulting in a median (IQR) of 20 (20-22) videos and 11 (9-14) minutes of total video time per patient. Patient outcomes were assessed retrospectively at the end of the study and, therefore, could not have influenced the imaging procedure.
In patients without a rash, we imaged the left volar forearm and left upper anterior chest as standard sites. In patients with a rash owing to suspected skin aGVHD, we imaged 1 to 2 rash sites and, if possible, 1 standard site (typically the forearm).
Confocal Video Analysis of Leukocyte-Endothelial Interactions
We quantified leukocyte-endothelial interactions in cutaneous microvessels by counting adherent and rolling leukocytes (Video). After aligning images to remove motion artifacts, 2 blinded raters (J.R.P. and Z.Z.) independently counted adherent and rolling leukocytes in each video based on a validated guideline.19 Raters viewed each video at least once. When raters counted different numbers, they reviewed the video together to achieve a consensus count. The sum of adherent and rolling leukocytes per patient was normalized to the total length of videos per patient, yielding the adherent and rolling leukocyte count per hour (A&R). Detection bias was minimized by ensuring that video acquisition was blinded to the resulting A&R and patient outcomes.
Video. Adherent and Rolling Leukocytes in Intact Cutaneous Microvasculature via Confocal Videomicroscopy.
Clinically Meaningful Threshold Selection for A&R
Before any outcome analysis of the 56 patients was completed, we selected a clinically meaningful, single universal threshold to divide patients into high and low A&R groups. Half of a population of 38 patients (including 36 from this study) suspected to have aGVHD had an A&R of 14 or higher, so we selected this as the threshold for high A&R. We also constructed linear and nonlinear models of A&R as a continuous variable in a sensitivity analysis.
rDRI Determination
The rDRI score is based on a look-up table of the cancer stage and response to treatment before HCT.3,20 We combined patients with low and intermediate risk (low rDRI) and patients with high and very high risk (high rDRI) according to Aziz et al.21 Additionally, we determined the composite score of DRI and HCT comorbidity index (DRCI),22 combining patients with very low, low, and intermediate 1 and 2 risk (intermediate DRCI) and patients with high and very high risk (high DRCI).
Clinical Outcome Determination
In December 2020, at least 6 months after the last patient imaging session, we reviewed the electronic medical records of each patient to determine whether the patient had died or relapsed. Relapse was determined from bone marrow biopsy. Outcomes were censored by the last clinic visit available in the record. Outcomes were calculated as the time from day of imaging to disease relapse (cumulative incidence of relapse [CIR]), disease relapse or death from any cause (relapse-free survival [RFS]), or death from any cause (OS).
Statistical Analysis
To lessen the influence of outlier values, A&R was transformed by cube root (eFigure 1 in the Supplement). Cox proportional hazards regression tested the association of A&R with relapse and/or death. Schoenfeld residual test supported using proportional hazards for all models. The Fine-Gray competing risk method23 with NRM as a competing risk was used to calculate CIR. The Kaplan-Meier method was used to calculate RFS and OS. Outcome associations were calculated in univariate, bivariable, and multivariable models of A&R, rDRI, and potential confounders of steroid treatment or presence of aGVHD. The likelihood-ratio χ2 test was used to evaluate the adequacy of each univariate model (A&R or rDRI) and bivariable model (A&R and rDRI).24 We used the Akaike information criterion to select the most informative A&R model for CIR, RFS, and OS. There were no missing data in this study. Power analysis is discussed in eMethods in the Supplement. Analyses were conducted using RStudio, version 2021.09.1 (RStudio), and tests were 2-sided with P < .05 considered statistically significant.
Results
Patient Characteristics and Follow-up
A total of 57 patients were imaged a median (IQR) of 40 (34-58) days after HCT for hematologic cancer. As standard of care, all patients had their underlying disease assessment at day 30 after transplant. One patient relapsed by imaging day and was excluded. On imaging day, all other patients were without clinical evidence of recurrence. This yielded 56 patients (median age, 59 years; 38 male, 18 female) for analyses. The most common diseases were acute myeloid leukemia (n = 26) and myelodysplastic syndrome (n = 14). Donors were matched unrelated (n = 35), matched related (n = 10), and haploidentical (n = 11).
The final medical record review was completed in May 2020, a median of 15 (IQR, 9-22; range, 6-38) months after HCT. At this review, 13 patients had relapsed and 14 died. Of the 14 patients who died, 8 died from complications of their underlying disease (relapse) and 6 died of other causes (NRM). No patients were lost to follow-up. Table 1 summarizes patient characteristics and follow-up, Table 2 summarizes patient outcomes, and eTable 1 in the Supplement lists each patient’s characteristics in chronological order of recruitment.
Table 2. Patient Outcomes.
| Variable | No. of patients | Total deaths | Deaths not associated with relapse (NRM) | Deaths associated with relapse | Total No. of relapses |
|---|---|---|---|---|---|
| Total | 56 | 14 | 6 | 8 | 13 |
| rDRI | |||||
| Low | 32 | 6 | 4 | 2 | 5 |
| High | 24 | 8 | 2 | 6 | 8 |
| Total | 56 | 14 | 6 | 8 | 13 |
| A&R | |||||
| None | 20 | 4a | 3 | 1 | 3 |
| Low | 35 | 5 | 3 | 2 | 4 |
| High | 21 | 9 | 3 | 6 | 9b |
| Total | 56 | 14 | 6 | 8 | 13 |
| No aGVHD | |||||
| Low A&R | 16 | 3 | 1 | 2 | 4 |
| High A&R | 4 | 3 | 0 | 3 | 3 |
| Total | 20 | 6 | 1 | 5 | 7 |
| Suspected aGVHD | |||||
| Low A&R | 19 | 2 | 2 | 0 | 0 |
| High A&R | 17 | 6 | 3 | 3 | 6 |
| Total | 36 | 8 | 5 | 3 | 6 |
Abbreviations: aGVHD, acute graft-vs-host disease; A&R, total number of adherent and rolling leukocytes per hour; NRM, nonrelapse mortality; rDRI, revised Disease Risk Index.
Of the 4 patients who died and had no A&R, 3 died of atypical NRM causes: vancomycin-resistant enterococcal bacteremia (rDRI: intermediate risk), deep vein thrombosis/pulmonary embolism (rDRI: intermediate risk), and post-transplant lymphoproliferative disorder (rDRI: very high risk). The fourth patient died of relapsed acute myeloid leukemia (rDRI: high risk).
Of 9 patients with high A&R who relapsed, 8 relapsed within 4 months after imaging.
In most patients, 2 body sites were imaged. In 5 patients, 1 site was imaged owing to limited time. In 4 patients with suspected skin aGVHD, 3 rash sites were imaged. Twenty patients had no symptoms of aGVHD, and 36 patients with suspected aGVHD were imaged as close as possible to symptom onset. Video time per patient did not differ considerably between those imaged with suspected aGVHD and those without symptoms (median video length, 12 vs 11 minutes; Wilcoxon rank-sum test, P = .95). No patients had severe aGVHD. Of the 36 patients with suspected aGVHD at the time of imaging, 21 had grade I and 4 had grade II aGVHD. Glucksberg grade25 was not listed for 11 patients, but medical record review did not reveal any severe symptoms consistent with grade III or IV aGVHD. Medical record review on day 100 after HCT confirmed that 22 patients had skin aGVHD (12 with Glucksberg grade ≤2), 10 had gut aGVHD (9 with Glucksberg grade ≤2), and 4 had both skin and gut aGVHD (all with Glucksberg grade ≤2).
Association of A&R and rDRI With Patient Outcomes
Associations between A&R and patient outcomes were calculated in a univariate model. Compared with low A&R, high A&R was associated with reduced OS (hazard ratio [HR], 3.24; 95% CI, 1.08-9.67; P = .04). An even stronger association of A&R with relapse was found (Table 3). Compared with patients with low A&R, those with high A&R had a statistically significantly higher CIR (HR, 4.59; 95% CI, 1.44-14.70; P = .01) and reduced RFS (HR, 3.37; 95% CI, 1.30-8.72; P = .01). There was a similar trend of reduced RFS and OS in the high A&R group when A&R was normalized to the absolute white blood cell count.
Table 3. Associations of A&R and rDRI With Patient Outcomes.
| Measure | No. of eventsa | High vs low A&R | High vs low rDRI | ||
|---|---|---|---|---|---|
| HR (95% CI) | P value | HR (95% CI) | P value | ||
| Univariate analysis (n = 56)b | |||||
| CIR | 13 | 4.59 (1.44-14.70) | .01 | 2.40 (0.79-7.32) | .12 |
| RFS | 18 | 3.37 (1.30-8.72) | .01 | 1.48 (0.59-3.72) | .41 |
| OS | 14 | 3.24 (1.08-9.67) | .04 | 1.90 (0.66-5.49) | .24 |
| Bivariable analysis (n = 56)c | |||||
| CIR | 13 | 4.24 (1.32-13.58) | .02 | 2.05 (0.67-6.33) | .21 |
| RFS | 18 | 3.29 (1.26-8.55) | .02 | 1.34 (0.53-3.39) | .54 |
| OS | 14 | 3.06 (1.02-9.19) | .05 | 1.69 (0.58-4.94) | .34 |
| Low and intermediate rDRI (n = 32)d | |||||
| CIR | 5 | 3.89 (0.71-21.30) | .12 | NA | NA |
| RFS | 9 | 2.01 (0.54-7.51) | .30 | NA | NA |
| OS | 6 | 2.34 (0.47-11.62) | .30 | NA | NA |
| High and very high rDRI (n = 24)e | |||||
| CIR | 8 | 4.75 (1.00-22.40) | .05 | NA | NA |
| RFS | 9 | 6.19 (1.25-30.59) | .03 | NA | NA |
| OS | 8 | 4.40 (0.88-21.88) | .07 | NA | NA |
Abbreviations: A&R, adherent and rolling leukocyte count per hour; CIR, cumulative incidence of relapse; HR, hazard ratio; NA, not applicable; OS, overall survival; rDRI, revised Disease Risk Index; RFS, relapse-free survival.
Number of events are the number of patients who relapsed (CIR), relapsed or died (RFS), or died (OS).
Univariate model of A&R or rDRI.
Bivariable model of A&R and rDRI.
A&R association with CIR, RFS, and OS in patients with low and intermediate rDRI.
A&R association with CIR, RFS, and OS in patients with high and very high rDRI.
The rDRI, which has been validated to predict relapse and OS after HCT, was also calculated. In the present population, the univariate associations of rDRI for CIR, RFS, and OS ranged from 1.48 to 2.40 but did not reach statistical significance (Table 3). Differences between this study population of 56 patients and the original study by Armand et al3 of 13 131 patients are summarized in eTable 2 in the Supplement. In bivariable analysis of A&R and rDRI, associations remained similar to the univariate analyses. Compared with low A&R, high A&R was associated with reduced OS (HR, 3.06; 95% CI, 1.02-9.19; P = .05; Figure, A) and RFS (HR, 3.29; 95% CI, 1.26-8.55; P = .02; Figure, B), and increased CIR (HR, 4.24; 95% CI, 1.32-13.58; P = .02; Figure, C) (Table 3). After correcting for baseline rDRI, patients with high A&R after HCT were 3.29 times more likely to relapse or die than patients with low A&R.
Figure. Kaplan-Meier Survival Analyses of Patients Based on Adherent and Rolling Leukocyte Counts per Hour (A&R).
Patients with high A&R counts relapsed at a higher rate (cumulative incidence of relapse) and had statistically significantly shorter relapse-free survival and overall survival (Table 3). Median survival of the group with high A&R was 215 days after imaging. In the subgroup of patients with low and intermediate revised Disease Risk Index (rDRI; n = 32; Table 3), A&R had a similar association with greater cumulative incidence of relapse and with shorter relapse-free survival and overall survival as in the study population. This was also true in the subgroup of patients with high and very high rDRI (n = 24; Table 3). HR indicates hazard ratio.
Use of corticosteroids is known to affect leukocyte peripheral concentration and endothelial interactions.26 On the imaging day, 15 patients (13 with suspected aGVHD and 2 without aGVHD) had been receiving topical (n = 5) and/or systemic (n = 10) corticosteroids for 1 to 45 days (IQR, 2 to 14 days). Similarly, aGVHD status may be associated with leukocyte-endothelial interactions.17 The compromised RFS, CIR, and OS were observed independently of status as suspected aGVHD or without aGVHD (eTable 3 and eFigure 2 in the Supplement) or steroid treatment (eTable 4 and eFigure 3 in the Supplement). The strength of the A&R association with outcomes was preserved when corrected for rDRI, steroid treatment, and aGVHD presence in bivariable models of A&R with each covariate, as well as in a single multivariable model (eTable 5 in the Supplement); A&R remained a better predictor for all 3 outcomes than rDRI in additional sensitivity analyses limiting outcome information to exactly 6 months follow-up (eTables 6 and 7 in the Supplement). Dichotomous DRCI assignment was identical to the rDRI assignment and therefore produced entirely identical results to the rDRI models for all analyses. Thus, leukocyte-endothelial interactions, quantified by the A&R, better predicted relapse and survival than rDRI.
A&R as an Adequate Predictor of Relapse and Death
To evaluate the adequacy of each univariate model (A&R or rDRI) and bivariable model (A&R and rDRI), the likelihood-ratio χ2 test was used as recommended by Califf et al.24 For RFS, univariate high vs low A&R had a χ2 of 6.44 (P = .01), and univariate high vs low rDRI had a χ2 of 0.67 (P = .41). The bivariable (A&R and rDRI) model had a χ2 of 6.81 (P = .03). Thus, when only A&R and rDRI are considered, without adjusting for other clinical variables, A&R accounts for 95% (6.44 of 6.81) of the prognostic information provided by combining A&R and rDRI to predict RFS. Looking at the adequacy of all assessed outcomes (OS, CIR, and RFS), the new imaging biomarker (A&R) accounted for 82% to 95% of the prognostic information (Table 4). By contrast, the best existing clinical predictor of rDRI accounted for 10% to 28% of the prognostic information in the same model. By this measure, A&R is a more adequate predictor of relapse or death than rDRI.
Table 4. Adequacy of Univariate and Bivariable Models for A&R and rDRI.
| Outcome | Univariate high vs low A&R | Univariate high vs low rDRI | Bivariable A&R and rDRI | Adequacy, % | ||||
|---|---|---|---|---|---|---|---|---|
| LR χ2a | P value | LR χ2a | P value | LR χ2a | P value | A&Rb | rDRIb | |
| CIR | 7.11 | .01 | 2.45 | .12 | 8.72 | .003 | 82 | 28 |
| RFS | 6.44 | .01 | 0.67 | .41 | 6.81 | .03 | 95 | 10 |
| OS | 4.69 | .03 | 1.43 | .23 | 5.63 | .06 | 83 | 25 |
Abbreviations: A&R, adherent and rolling leukocyte count per hour; CIR, cumulative incidence of relapse; LR, likelihood ratio; OS, overall survival; rDRI, revised Disease Risk Index; RFS, relapse-free survival.
LR χ2 test based on a Cox proportional hazards survival model.
Prognostic information of univariate marker relative to a bivariable A&R and rDRI model (univariate LR χ2 test/bivariable LR χ2 test).
To test whether thresholding distorted the associations, a sensitivity analysis was conducted. The AIC of the dichotomous analysis was compared with linear and nonlinear models of A&R as a continuous variable. Based on the AIC, the best A&R model for CIR, RFS, and OS was the dichotomous high vs low A&R model (eTable 8 in the Supplement). In the continuous models, the probability of dying within 6 months after imaging monotonically decreased with increasing A&R value (eFigure 4 in the Supplement). The higher the A&R, the more likely the patient will relapse or die. The rank-ordered rDRI model achieved similar AICs as the dichotomous rDRI model, indicating similar model quality for the data.
Discussion
We discovered that leukocyte-endothelial interactions measured within 100 days after HCT were associated with important patient outcomes. In multivariable models, the proposed marker of A&R had a stronger association with relapse, RFS, and OS after HCT than rDRI. These associations were preserved after correcting for possible confounders, steroid treatment, and aGVHD status. Thus, leukocyte-endothelial interactions, quantified by the A&R, better predicted relapse and survival than rDRI.
Although prognostically weaker than A&R, the rDRI association with OS in the present patient population (HR, 1.69) was consistent with the original report by Armand et al (HRs, 1.46-2.97)3 but with statistical significance lessened by the smaller sample size. Furthermore, the present study had fewer patients with low or intermediate rDRI (58% vs 76%), more male patients (68% vs 58%), and more patients with an HCT comorbidity index of 1 or 2 (36% vs 28%) or 3 or more (41% vs 30%). Although aGVHD status was not reported by Armand et al, the present base in the Department of Dermatology at Vanderbilt University Medical Center and the US Department of Veterans Affairs in Nashville, Tennessee, likely selects for mild cutaneous aGVHD. These population characteristics may limit generalizability to settings with less selected groups.
Current assessments for pre-HCT risk are static and do not change after HCT, limiting their value for disease management. Thus, a dynamic post-HCT marker is needed to monitor patient status and predict relapse. Leukocyte-endothelial interactions can be assessed noninvasively and monitored after HCT. Thus, regardless of pre-HCT risk, leukocyte-endothelial interactions may help clinicians to identify and monitor patients at risk for relapse and death, and to weigh the risks and benefits of more intensive treatments after HCT. For example, patients at higher risk for relapse could be enrolled in experimental studies.
We do not yet know what mechanism links higher leukocyte-endothelial interactions with subsequent relapse of the underlying hematologic cancer; A&R was more strongly associated with relapse and RFS than with OS. Higher interactions could be because of changes in endothelial activation, which can be indirectly scored with the Endothelial Activation and Stress Index (EASIX). However, EASIX has been more strongly associated with NRM than relapse.5 Higher interactions could also reflect improper leukocyte homing to the vasculature. Alternatively, we may have visualized leukemic cells, which express adhesion molecules and easily bind endothelial cells. Similar to other videomicroscopy studies of human microcirculation,6,7 we did not identify the type of leukocytes visualized. Currently, live-cell labeling strategies are not compatible with clinical use of any FDA-approved confocal system.
This study was designed to align with our institutional practices, which do not involve collecting samples for prognostic biomarkers in routine patient care. Future validation studies should collect blood-based biomarkers (eg, Mount Sinai Acute GVHD International Consortium,4 EASIX5) for direct comparison with the A&R.
Limitations
This study is limited by the sample size of 56 patients and lack of an external validation population. With only 13 and 14 events (for relapse and death, respectively), multivariable models risk overfitting. We therefore recommend that validation studies use the bivariable (rather than multivariable) model results for effect-size estimation.
Conclusions
In this cohort study, leukocyte-endothelial interactions assessed noninvasively by confocal videomicroscopy had statistically significant associations with relapse, relapse-free survival, and overall survival. This study identified a promising systemic disease biomarker from leukocyte-endothelial interactions imaged by noninvasive confocal videomicroscopy of skin microvessels with an FDA-approved patient care device. This approach could be used to directly quantify intact immunological processes at the patient bedside in many clinical applications and could reveal novel cutaneous biomarkers that inform systemic treatment decisions.
eFigure 1. Histograms of (A) A&R counts and (B) cube root [A&R + 1] counts of all study patients (N=56)
eMethods. Power Calculation
eTable 1. Each patient’s characteristics in chronological order of recruitment
eTable 2. Summary of differences between our study population of 56 patients and Armand’s original study of 13 131 patients
eTable 3. Univariate Associations of A&R in Subgroups of Patients Suspected of aGVHD and No aGVHD
eFigure 2. A&R Counts in Patients with Suspected aGVHD and No aGVHD
eTable 4. Univariate Associations of A&R in Patients Taking or Not Taking Steroids
eFigure 3. A&R Counts in Patients Taking and Not Taking Steroids Before Imaging
eTable 5. Associations of A&R and Outcomes After Correcting for rDRI and Potential Confounders
eTable 6. Associations of A&R and rDRI with outcomes, with follow-up capped at 6 months
eTable 7. Adequacy of Univariate and Bivariable Models for A&R and rDRI, with follow-up capped at 6 months
eTable 8. Comparison of AIC and BIC for Various A&R and rDRI Model Forms
eFigure 4. Probability of Survival at 3, 6, and 9 Months
eReference
References
- 1.Bair SM, Brandstadter JD, Ayers EC, Stadtmauer EA. Hematopoietic stem cell transplantation for blood cancers in the era of precision medicine and immunotherapy. Cancer. 2020;126(9):1837-1855. doi: 10.1002/cncr.32659 [DOI] [PubMed] [Google Scholar]
- 2.D’Souza A, Fretham C, Lee SJ, et al. Current use of and trends in hematopoietic cell transplantation in the United States. Biol Blood Marrow Transplant. 2020;26(8):e177-e182. doi: 10.1016/j.bbmt.2020.04.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Armand P, Kim HT, Logan BR, et al. Validation and refinement of the Disease Risk Index for allogeneic stem cell transplantation. Blood. 2014;123(23):3664-3671. doi: 10.1182/blood-2014-01-552984 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Major-Monfried H, Renteria AS, Pawarode A, et al. MAGIC biomarkers predict long-term outcomes for steroid-resistant acute GVHD. Blood. 2018;131(25):2846-2855. doi: 10.1182/blood-2018-01-822957 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Shouval R, Fein JA, Shouval A, et al. External validation and comparison of multiple prognostic scores in allogeneic hematopoietic stem cell transplantation. Blood Adv. 2019;3(12):1881-1890. doi: 10.1182/bloodadvances.2019032268 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Fabian-Jessing BK, Massey MJ, Filbin MR, et al. ; ProCESS Investigators . In vivo quantification of rolling and adhered leukocytes in human sepsis. Crit Care. 2018;22(1):240. doi: 10.1186/s13054-018-2173-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Kirveskari J, Vesaluoma MH, Moilanen JAO, et al. A novel non-invasive, in vivo technique for the quantification of leukocyte rolling and extravasation at sites of inflammation in human patients. Nat Med. 2001;7(3):376-379. doi: 10.1038/85538 [DOI] [PubMed] [Google Scholar]
- 8.Carman CV, Martinelli R. T lymphocyte-endothelial interactions: emerging understanding of trafficking and antigen-specific immunity. Front Immunol. 2015;6:603. doi: 10.3389/fimmu.2015.00603 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Granger DN, Kubes P. The microcirculation and inflammation: modulation of leukocyte-endothelial cell adhesion. J Leukoc Biol. 1994;55(5):662-675. doi: 10.1002/jlb.55.5.662 [DOI] [PubMed] [Google Scholar]
- 10.Joseph A, Chu CJ, Feng G, Dholakia K, Schallek J. Label-free imaging of immune cell dynamics in the living retina using adaptive optics. Elife. 2020;9:1-14. doi: 10.7554/eLife.60547 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Huang L, McKay GN, Durr NJ. A deep learning bidirectional temporal tracking algorithm for automated blood cell counting from non-invasive capillaroscopy videos. arXiv. Preprint published online November 26, 2020. Updated April 12, 2021. doi: 10.48550/arXiv.2011.13371 [DOI]
- 12.Rajadhyaksha M, Grossman M, Esterowitz D, Webb RH, Anderson RR. In vivo confocal scanning laser microscopy of human skin: melanin provides strong contrast. J Invest Dermatol. 1995;104(6):946-952. doi: 10.1111/1523-1747.ep12606215 [DOI] [PubMed] [Google Scholar]
- 13.Liu H, Xuan L, Lin R, et al. A new pre-emptive TKIs strategy for preventing relapse based on BCR/ABL monitoring for Ph+ALL undergoing allo-HCT: a prospective clinical cohort study. Leukemia. 2021;35(7):2054-2063. doi: 10.1038/s41375-020-01090-4 [DOI] [PubMed] [Google Scholar]
- 14.Ardigo M, Longo C, Gonzalez S; International Confocal Working Group Inflammatory Skin Diseases Project . Multicentre study on inflammatory skin diseases from The International Confocal Working Group: specific confocal microscopy features and an algorithmic method of diagnosis. Br J Dermatol. 2016;175(2):364-374. doi: 10.1111/bjd.14516 [DOI] [PubMed] [Google Scholar]
- 15.Saknite I, Gill M, Alessi-Fox C, et al. Features of cutaneous acute graft-versus-host disease by reflectance confocal microscopy. Br J Dermatol. 2019;181(4):829-831. doi: 10.1111/bjd.17921 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Arora M, Cutler CS, Jagasia MH, et al. Late acute and chronic graft-versus-host disease after allogeneic hematopoietic cell transplantation. Biol Blood Marrow Transplant. 2016;22(3):449-455. doi: 10.1016/j.bbmt.2015.10.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Saknite I, Byrne MT, Jagasia M, Tkaczyk ER. Noninvasive microscopic imaging reveals increased leukocyte adhesion and rolling in skin of acute graft-versus-host disease patients compared to post-transplant controls. Blood. 2019;134(suppl 1):4533. doi: 10.1182/blood-2019-123795 [DOI] [Google Scholar]
- 18.Saknite I, Zhao Z, Patrinely JR, Byrne M, Jagasia M, Tkaczyk ER. Individual cell motion in healthy human skin microvasculature by reflectance confocal video microscopy. Microcirculation. 2020;27(6):e12621. doi: 10.1111/micc.12621 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zhao Z, Patrinely JR, Saknite I, Byrne M, Tkaczyk ER. Guideline for in vivo assessment of adherent and rolling leukocytes in human skin microvasculature via reflectance confocal videomicroscopy. Microcirculation. 2021;28(8):e12725. doi: 10.1111/micc.12725 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Armand P, Gibson CJ, Cutler C, et al. A disease risk index for patients undergoing allogeneic stem cell transplantation. Blood. 2012;120(4):905-913. doi: 10.1182/blood-2012-03-418202 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Aziz MD, Shah J, Kapoor U, et al. Disease risk and GVHD biomarkers can stratify patients for risk of relapse and nonrelapse mortality post hematopoietic cell transplant. Leukemia. 2020;34(7):1898-1906. doi: 10.1038/s41375-020-0726-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Bejanyan N, Brunstein CG, Cao Q, et al. Predictive value of disease risk comorbidity index for overall survival after allogeneic hematopoietic transplantation. Blood Adv. 2019;3(3):230-236. doi: 10.1182/bloodadvances.2018018549 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Gray RJ. A class of κ-sample tests for comparing the cumulative incidence of a competing risk. Ann Stat. 1988;16(3):1141-1154. doi: 10.1214/aos/1176350951 [DOI] [Google Scholar]
- 24.Califf RM, Phillips HR III, Hindman MC, et al. Prognostic value of a coronary artery jeopardy score. J Am Coll Cardiol. 1985;5(5):1055-1063. doi: 10.1016/S0735-1097(85)80005-X [DOI] [PubMed] [Google Scholar]
- 25.Cahn JY, Klein JP, Lee SJ, et al. ; Société Française de Greffe de Moëlle et Thérapie Cellulaire; Dana Farber Cancer Institute; International Bone Marrow Transplant Registry . Prospective evaluation of 2 acute graft-versus-host (GVHD) grading systems: a joint Société Française de Greffe de Moëlle et Thérapie Cellulaire (SFGM-TC), Dana Farber Cancer Institute (DFCI), and International Bone Marrow Transplant Registry (IBMTR) prospective study. Blood. 2005;106(4):1495-1500. doi: 10.1182/blood-2004-11-4557 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Poetker DM, Reh DD. A comprehensive review of the adverse effects of systemic corticosteroids. Otolaryngol Clin North Am. 2010;43(4):753-768. doi: 10.1016/j.otc.2010.04.003 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eFigure 1. Histograms of (A) A&R counts and (B) cube root [A&R + 1] counts of all study patients (N=56)
eMethods. Power Calculation
eTable 1. Each patient’s characteristics in chronological order of recruitment
eTable 2. Summary of differences between our study population of 56 patients and Armand’s original study of 13 131 patients
eTable 3. Univariate Associations of A&R in Subgroups of Patients Suspected of aGVHD and No aGVHD
eFigure 2. A&R Counts in Patients with Suspected aGVHD and No aGVHD
eTable 4. Univariate Associations of A&R in Patients Taking or Not Taking Steroids
eFigure 3. A&R Counts in Patients Taking and Not Taking Steroids Before Imaging
eTable 5. Associations of A&R and Outcomes After Correcting for rDRI and Potential Confounders
eTable 6. Associations of A&R and rDRI with outcomes, with follow-up capped at 6 months
eTable 7. Adequacy of Univariate and Bivariable Models for A&R and rDRI, with follow-up capped at 6 months
eTable 8. Comparison of AIC and BIC for Various A&R and rDRI Model Forms
eFigure 4. Probability of Survival at 3, 6, and 9 Months
eReference