Examination of DNAm PhenoAge as an epigenetic biomarker for perioperative risk in adult spinal deformity surgeries

Abstract

Epigenetic changes, such as DNA methylation (DNAm), offer a measure of biological age distinct from chronological age. DNAm PhenoAge is one such biomarker that is more strongly related to morbidity, mortality, and physical function than chronological age. More accurate risk stratification methods are needed for ASD surgeries, where complications remain difficult to predict with an increasingly aged population. A multicenter ASD registry was queried. DNAm PhenoAge was calculated as per Levine et al. (6). Multivariable logistic regression examined the associations of DNAm PhenoAge and chronological age with perioperative adverse events (AE). The relative improvements in model discrimination, fit, and classification performance were compared. Adjusted odds ratios compared the risk of 55 versus 75 years for each age metric. Laboratory data were available for 200 patients. Mean DNAm PhenoAge was lower than chronological (DNAm PhenoAge, 53.7 ± 18.1; chronological, 61.1 ± 15.4; p < 0.001; 95% CI, 6.3–8.5). DNAm PhenoAge models demonstrated numerically higher discrimination for all outcomes examined (AUC range, 0.701–0.823 vs. 0.671–0.767), although no DeLong comparisons were statistically significant (all p > 0.05). Model fit consistently favored DNAm PhenoAge (ΔAIC, 4.41–7.08), with NRI ranging from 0.198 to 0.500. For most adverse events, DNAm PhenoAge demonstrated adjusted odds ratios that were comparable to or greater than those observed for chronological age. In this exploratory study, DNAm PhenoAge demonstrated associations with several postoperative adverse events. Although differences in discrimination were not statistically significant, improvements were consistently observed across multiple complementary measures of model performance. The generally consistent direction of improved model performance suggests that DNAm PhenoAge warrants further study as a candidate biomarker for perioperative risk stratification (level of evidence, 3).

Introduction

Adult spinal deformity (ASD) is increasing in prevalence with an aging population and symptomatic patients often seek surgical care [1]. The rates of complications and reoperations may make ASD surgery unsustainable in a cost-constrained healthcare economy, as they reduce value through higher costs and diminished health-related quality of life (HRQOL) [2]. Thus, there is a pressing need for improved risk stratification, decision-making, and surgical optimization in ASD care [3]. Numerous studies have reported risk factors for poor outcomes, low health-related quality-of-life (HRQOL), and reoperation. These studies typically rely on easily accessible clinical variables, including chronological age; however, these factors often demonstrate limited predictive performance for individual outcomes [4, 5]. The inability to accurately predict outcomes in ASD surgery may be related to unmeasured variables, rather than insufficient sample sizes.

One such unmeasured factor may be biological age, a measure of time-dependent cellular decline and accumulation of health deficits not measured by chronological age [6, 7]. Epigenetic modifications, such as DNA methylation (DNAm) and demethylation, are quantifiable changes used to estimate biological age [8]. Measures of DNAm obtained from peripheral blood samples are associated with a greater risk of complications in ASD surgeries and death after hip fracture [9, 10]. The Levine DNAm PhenoAge clock is a biomarker more strongly associated with comorbidity and mortality compared to chronological age. It can be calculated from standard peripheral blood samples without the need for DNA isolation and analyses [6]. This makes DNAm PhenoAge readily available for clinical research and possible deployment.

The purpose of this study was to examine the relationship between DNAm PhenoAge and chronological age with complications in patients undergoing surgery for ASD. We hypothesize that DNAm PhenoAge will demonstrate stronger discriminatory power than chronological age for predicting perioperative complications following ASD surgery.

Methods

Data source and patient cohort

A multicenter registry (Clinicaltrials.gov: NCT04194138) of ASD patients was queried for patients enrolled from 2018 to 2021 with laboratory data for DNAm PhenoAge calculation. Institutional review board approval was obtained at all sites. All patients were at least 18 years of age and had a diagnosis of adult congenital, degenerative, idiopathic, or iatrogenic thoracolumbar spinal deformity with plan for operative treatment. Exclusion criteria included active spinal infection/neoplasm, acute trauma, neuromuscular/syndromic scoliosis, inflammatory arthritis/autoimmune diseases, prisoners, and pregnancy or immediate plans to become pregnant.

Preoperative laboratory collection was not standardized and only patients with available data for DNAm PhenoAge calculation were included thereby performing a complete case analysis. DNAm PhenoAge was calculated according to the method of Levine et al. [6]. A mortality score was calculated from a weighted summary of age and laboratory data (Table 1), and a DNAm PhenoAge was derived from this mortality score. Standardized case report forms collected demographic, clinical, and radiographic data. Comorbid conditions were recorded and the Charlson Comorbidity Index (CCI) calculated. The Edmonton Frailty Score measured frailty. American Spinal Injury Association (ASIA) motor exams were recorded. Surgical data included estimated blood loss (EBL, mL) and whether a three-column osteotomy (3CO) was performed. General (Patient-Reported Outcomes Measurement Information System (PROMIS)) and disease-specific patient-reported outcome scores were collected before surgery.

Table 1.

Demographic, patient-reported outcome measure, laboratory, and radiographic data presented with missingness of imputed values

Characteristic	N = 2001	Missing
Chronological age, yr		0
Mean (SD)	61.1 (15.4)
Median (Q1, Q3)	66.5 (55.5, 71.2)
DNAm PhenoAge, yr		0
Mean (SD)	54.9 (19.1)
Median (Q1, Q3)	58.9 (46.1, 67.0)
Sex		0
Female	144 (72.0%)
Male	56 (28.0%)
Race		0
Asian	5 (2.5%)
Black/African American	4 (2.0%)
Other	3 (1.5%)
White	188 (94.0%)
Height, cm	163.0 (157.5, 172.0)	8
Weight, kg	72.3 (61.2, 85.7)	8
Body mass index, kg/m2	27.1 (23.6, 30.6)	8
Hand-grip strength	54.0 (43.0, 64.0)	8
Edmonton frailty score	3.0 (2.0, 5.0)	8
Charlson comorbidity index	0.0 (0.0, 1.5)	8
Prior thoracolumbar fusion	99 (49.7%)	1
Preoperative C7 sagittal vertical axis	58.1 (14.2, 102.8)	16
Cardiac comorbid condition	10 (5.2%)	8
PROMIS anxiety	55.4 (50.6, 61.5)	10
Preexisting lower-extremity neurological deficit	68 (42.8%)	41
Estimated blood loss, mL		0
Median (Q1, Q3)	1000 (600, 1600)
Min to max	100 to 6000
Operative time, min		20
Median (Q1, Q3)	479 (399, 555)
Min to max	117 to 1301
Pedicle subtraction osteotomy	34 (17.0%)	0
Vertebral column resection	8 (4.0%)	0
Three-column osteotomy	41 (20.5%)	0
Discharge disposition		32
1	108 (64.3%)
2	14 (8.3%)
3	46 (27.4%)
Any adverse event	103 (51.5%)	0
Medical adverse event	60 (30.0%)	0
Cardiac adverse event	36 (18.0%)	0
Systemic infection	12 (6.0%)	0
Readmission	34 (17.0%)	0
Neurological deficit	28 (14.0%)	0

^*A component of the DNAm PhenoAge calculator

¹n (%); median (Q1, Q3)

Sites were queried for missing data on an ongoing basis for quality assurance. Prior to analysis, the dataset was examined for missing data. No covariate had missingness more than 10% and missing covariate data were imputed using predictive mean matching (single imputation, ten iterations). As this is an exploratory retrospective study, our sample size was limited by the enrollment in the registry, and no a priori sample size estimation was performed.

Complications

Intraoperative data and perioperative data were collected on standardized case report forms and complications were recorded. Reported complications were adjudicated by a panel of surgeons independent from the surgery. Perioperative complications were categorized by acuity, affected body system and by severity of treatment required [11]. Only complication types with more than 10 observed events were included in the multivariable analyses to limit model instability.

Statistical analyses

Demographic data were reported using standard descriptive statistics. Correlations between DNAm PhenoAge and baseline demographic and radiographic data were examined with Pearson and Spearman correlations as appropriate. Mean chronological age and DNAm PhenoAge were compared between patients that experienced a complication and those who did not.

Multivariate logistic regression examined the relationship between each complication category and both chronological age and DNAm PhenoAge. Potential confounders were chosen by clinical evidence. Ages were modeled as nonlinear variables using restricted cubic splines with three knots. For each complication category, adjusted odds ratios with 95% confidence intervals were calculated by comparing the odds of a complication at age 75 versus age 55, separately for DNAm PhenoAge and chronological age. Model discrimination was described using the area under the curve (AUC) and differences were compared with DeLong’s p-value. Model fit was measured by the Akaike information criterion (AIC). The continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) compared risk reclassification between models.

Statistical analyses were performed using R (v4.1.2, R Project for Statistical Computing). Statistical significance was defined by a two-sided p < 0.05 and confidence intervals not including the null where applicable. As this was an exploratory study, corrections for multiple comparisons were not performed. In lieu of post hoc power analysis, we provide 95% confidence intervals to convey precision estimates.

Results

Demographics

Of the five hundred and six patients enrolled in the registry, two hundred patients (40%) were eligible and included in these analyses (Table 1). The cohort was predominantly female (144, 72%) with a median chronological age of 66.5 years (IQR, 55.5–71.3) and a median DNAm PhenoAge of 57.9 years (45.5–65.4). There was a strong correlation between chronological and DNAm PhenoAge (r = 0.9; Fig. 1). The median Edmonton frailty score was 2 (IQR, 2–5) indicating fewer than 25% of the cohort would be categorized as “Frail”. The Edmonton Frailty Score and DNAm PhenoAge were weakly correlated (r = 0.39, p < 0.001). Nearly half had prior thoracolumbar fusions (99, 49.5%), and 42 patients (21%) underwent a three-column osteotomy (34 pedicle subtraction osteotomies, eight vertebral column resections).

Fig. 1.

DNAm PhenoAge showed a strong relationship (r = 0.9) with chronological age

Complications

Over half (103, 52%) of the patients experienced some adverse event. Approximately one-third (60, 30%) of patients experienced medical adverse events. Subjects experiencing adverse events were generally older by both chronological age and DNAm PhenoAge. Patients who experienced a new spine-related neurological deficit (motor or sensory) had significantly older DNAm PhenoAge values (p = 0.015), whereas chronological age did not differ significantly between groups. Six adverse event (AE) categories met criteria for multivariable modeling: any AE (n = 103), any medical AE (n = 60), any cardiac AE (n = 36), any systemic infection (n = 12), unplanned readmission (n = 34), and new neurological deficit (n = 28).

Evaluation of DNAm PhenoAge

Model performances for each outcome and age metric are summarized in Table 2, with probability plots found in Fig. 2. Compared with chronological age, DNAm PhenoAge models demonstrated numerically higher AUCs for any adverse event (0.701 vs. 0.671), systemic infection (0.823 vs. 0.767), readmission (0.778 vs. 0.730), and neurological deficit (0.716 vs. 0.693). AUC differences ranged from −0.002 to 0.056. None of the paired ROC comparisons reached statistical significance using DeLong testing (all p > 0.05).

Table 2.

Model performance metrics for adverse events (AE)

Outcome	AUC, chronological	AUC, DNAm PhenoAge	ΔAUC	DeLong p	ΔAIC	NRI	IDI
Any AE	0.671 (0.595, 0.747)	0.701 (0.628, 0.773)	0.030	0.121	7.082	0.321	0.031
Medical AE	0.652 (0.573, 0.732)	0.65 (0.571, 0.729)	−0.002	0.921	−0.336	−0.252	−0.002
Cardiac AE	0.649 (0.56, 0.739)	0.66 (0.571, 0.749)	0.011	0.684	−1.075	−0.289	−0.010
Systemic infection AE	0.767 (0.623, 0.911)	0.823 (0.695, 0.951)	0.056	0.219	4.996	0.500	0.032
Readmission	0.73 (0.645, 0.815)	0.778 (0.702, 0.854)	0.048	0.075	6.416	0.277	0.036
New neurological deficit	0.693 (0.593, 0.792)	0.716 (0.618, 0.813)	0.023	0.685	4.407	0.198	0.027

AUC area under the curve (ΔAUC measured as DNAm PhenoAge AUC minus chronological AUC), NRI net reclassification improvement, IDI integrated discrimination improvement

Fig. 2.

Adjusted probability plots for each adverse event by chronological and DNAm PhenoAge metrics

DNAm PhenoAge models demonstrated lower AIC values than chronological age models for any adverse event (ΔAIC = 7.08), systemic infection (ΔAIC = 5.00), readmission (ΔAIC = 6.42), and neurological deficit (ΔAIC = 4.41). Net reclassification improvement ranged from 0.198 to 0.500 for these outcomes when comparing DNAm PhenoAge with chronological age, with corresponding IDI values ranging from 0.027 to 0.036.

Figure 3 compares the adjusted odds of each adverse event at age 75 versus age 55, for both chronological age and DNAm PhenoAge as the age metric. For most AEs, increasing phenotypic age demonstrated effect estimates (adjusted odds ratio, aOR) that were comparable to or greater than those observed for chronological age. The largest differences were observed for any adverse event (aOR, 2.15 vs. 1.79), systemic infection (aOR, 2.69 vs. 1.59), readmission (aOR, 2.45 vs. 2.13), and new neurological deficit (aOR, 1.89 vs. 0.83). Associations with medical and cardiac adverse events were similar between the two metrics. Confidence intervals are wide and overlapped for all comparisons.

Fig. 3.

Forest plot of adjusted odds of adverse events associated with a 20-year increase in age (75 vs. 55 years) using chronological age and phenotypic age as the age metric

Discussion

In this exploratory study, we find preliminary evidence supporting further study of DNAm PhenoAge as a biomarker of physiological vulnerability to adverse events following ASD surgery. DNAm PhenoAge improved model discrimination and fit (AUC and AIC) over chronological age in four of six adverse event models, though DeLong’s test did not reach significance for any outcome (p = 0.075–0.921). This study was not powered to detect the modest incremental gains expected from a single biomarker, particularly for outcomes with few events. The consistent direction of improvement in discrimination, model fit, and reclassification (AUC, AIC, NRI and IDI) supports DNAm PhenoAge as a candidate biomarker of surgical vulnerability.

While the probability plots were similar for most outcomes, a notable difference was observed for new neurological deficits. Increasing chronological age was associated with a lower predicted probability of new neurological deficit, whereas increasing DNAm PhenoAge was associated with a higher predicted probability (aOR, 1.89 vs. 0.83). One possible explanation is that chronological age influences procedural selection, with surgeons choosing less aggressive surgeries in older patients, while DNAm PhenoAge may better reflect an underlying biological susceptibility to neurological injury. This latter hypothesis is consistent with emerging evidence implicating biological age, tissue properties, and immune function in spinal cord injury [12].

Only a moderate correlation was observed between the Edmonton Frailty Scale and DNAm PhenoAge, suggesting that clinical frailty and biological aging are distinct but overlapping dimensions of patient vulnerability. This distinction may be particularly relevant in adult spinal deformity patients when disease-related disability, frailty, and comorbidity may coexist [13]. Levine emphasized the need for a continuous measure, noting that dichotomization of disease states lacks discrimination of both pre-disease and post-diagnosis states, leading us away from an accurate assessment of “system integrity” [14].

DNA methylation has been explored as a biomarker in adult spinal deformity surgery, although no association was observed between epigenetic age and a composite complication outcome in a small cohort of 83 patients [9]. Given the small sample size, this is not unexpected for a modest improvement in prediction from a single biomarker. Those patients sustaining complications had shorter telomere lengths than those without complications. While different from DNA methylation, both telomere length and DNA methylation are measures of biological aging. Taken together, the results of Safaee et al. [9] and those presented here support the further exploration of epigenetic measures of cellular “health” as biomarkers for perioperative risk stratification. We chose DNAm PhenoAge, rather than other geroscience biomarkers (e.g., DunedinPACE, GrimAge), because its estimate of accumulated physiological aging is conceptually more aligned with the physiological reserve available at the time of surgery than measures of the current pace of aging or mortality risk. We suspect that different biomarkers or epigenetic clocks are needed for different clinical outcomes, analogous to the distinct purposes served by DNAm PhenoAge, DunedinPACE, and GrimAge.

It is important to note that the relationships observed between complications and both chronological age and DNAm PhenoAge are relatively weak with wide confidence intervals. These findings should be viewed as hypothesis generating and evidence of the need for further investigation, rather than as “truth” or clinically irrelevant. It is important to consider that these variables are one component of a complex, multidimensional, biological system and other variables are required for prediction model creation, including surgical details. Furthermore, variance and stochasticity almost certainly exist from the point of DNA transcription to protein expression and the multitude of other subsequent biochemical interactions that happen beyond the epigenome. For that reason, it is likely that the evolution of biomarkers for risk prediction will go beyond epigenetic measures. We believe, however, that this is an important first step toward the goal of improving risk stratification with evidence of a biomarker consisting of standard laboratory measures that may be more useful than the commonly used chronological age.

This study has important limitations. First, only patients with complete laboratory data to calculate DNAm PhenoAge were included, which may introduce selection bias and limit the generalizability of the findings. Additionally, the cohort was predominantly older adults undergoing complex ASD surgery, with relatively few younger or biologically healthy patients. The resulting distribution may have limited the precision of the estimated associations across the full range of chronological and DNAm PhenoAge. Second, several adverse event categories had relatively few events, resulting in imprecise estimates, increased risk of model instability, and increased risk of chance findings. However, we believe examining more specific complication categories is an improvement over prior works combining complications into composites such as “major” and “minor.” Composite outcomes make external validation difficult or impossible, though the purpose of this study was to examine for potential benefits to using a biomarker for biological age, as we take a systems biology approach to ASD risk stratification [15]. We have examined ASD patients, in part, because of the high complication rates. Whether these findings will translate to other surgical domains is unknown, though ASD patients exhibit a postoperative immunocompromised state similar to other episodes after major trauma [16]. Similarly, advanced DNAm PhenoAge is associated with increased odds of mortality after geriatric hip fracture [10]. Finally, this study did not develop or validate clinical prediction models. We have only investigated the potential utility of DNAm PhenoAge as a biomarker to better predict complications related to physiological decline. The findings support further investigation of DNAm PhenoAge as a potential adjunct to perioperative risk assessment, including external validation in independent cohorts and evaluation alongside established clinical and surgical risk factors.

The work of Levine et al. [6] has paved the way for near- and long-term improvements in our risk stratification of surgery patients by developing phenotypic age. In this exploratory study, DNAm PhenoAge demonstrated associations with several postoperative adverse events that differed from those observed with chronological age. The generally consistent direction of improved model performance suggests that DNAm PhenoAge warrants further study as a candidate biomarker. Other reported risk factors have not consistently translated into meaningful reductions in complication rates, and biomarkers such as biological age may represent one of several incremental additions needed to advance risk prediction in ASD surgery. Further work is needed to determine whether DNAm PhenoAge provides clinically meaningful incremental value for perioperative risk stratification and whether it identifies modifiable pathways that could reduce surgical risk.

Data availability

The participants of this study did not give written consent for their data to be shared publicly, so due to the sensitive nature of the research supporting data is not available.

Declarations

Conflict of interest

Dr. Kelly reported receiving nonfinancial support from AO Spine and personal fees from Wolters Kluwer outside the submitted work. Dr. Hills reports no conflicts of interest. Dr. Smith reported receiving grants from DePuy Synthes/ISSGF during the conduct of the study; personal fees from Highridge, Globus/NuVasive, Medtronic, Carlsmed, Cerapedics, and Orthofix/SeaSpine; being a stock owner in Carlsmed, Alphatec, and Globus/NuVasive; and receiving grants from Orthofix/SeaSpine, the Neurosurgery Research & Education Foundation, and AO Spine, outside the submitted work. Dr. Lenke reported receiving grants from Scoliosis Spine Research, EOS, the Setting Scoliosis Straight Foundation, AOSpine, and the ISSG, receiving royalties and consulting fees from Medtronic, consulting fees from Abryx and Acuity Surgical, and nonfinancial support from Broadwater and the SRS for trips/travel, outside the submitted work. Dr. Kim reported receiving personal fees from Zimmer Biomet, K2M/Stryker, and Acuity Surgical, and grants from the ISSGF, outside the submitted work; having patents for NuVasive and Aspen Medical with royalties paid; and receiving payments to his institution from SI-BONE and AO Spine (fellowship support). Dr. Bess reported receiving grants from Medtronic, Globus, NuVasive, Stryker, Carlsmed, and the ISSGF, during the conduct of the study; and receiving grants from DePuy Synthes, SeaSpine, and the ISSGF, and personal fees from ATEC, outside the submitted work. B. Line reported receiving personal fees from the ISSG outside the submitted work. Dr. V. Lafage reported receiving consulting fees from Alphatec, Globus Medical, and Mainstay Medical; royalties from NuVasive; lecture fees from Johnson & Johnson, Stryker, and Implanet; being on the executive committee for the ISSG; and being a committee member for the Scoliosis Research Society (SRS), outside the submitted work. Mr. R. Lafage reported receiving personal fees from Carlsmed, outside the submitted work. Dr. Klineberg reported receiving consulting fees from DePuy Synthes, Medtronic, SeaSpine, SI-BONE, and Agnovos; consulting and royalty fees from Stryker Spine; having stock ownership in MMI and Relateable; receiving speaker fees from AO Spine North America (AONA Spine); chairing the board of directors for the International Meeting on Advanced Spine Techniques and AONA Spine; and being on the board of directors for the SRS, outside the submitted work. Dr. Pellisé reports consulting fees from Medtronic and DePuy Spine. Dr. Kebaish reported receiving consulting fees from DePuy Synthes and royalties from Stryker, Orthofix, and SpineCraft, outside the submitted work. Dr. Gupta reported receiving royalties from Innomed; consultant, advisory, and travel fees from Medtronic; royalties and consultant and travel fees from Globus and DePuy; owning stock and receiving travel fees from SMAIO; owning stock in Johnson & Johnson; being on the board of directors of the Twentieth Century Orthopaedic Association; having travel expenses paid for by Broadwater; receiving honoraria and travel expenses from Yale; and receiving travel expenses from and being on the board of directors of the SRS, outside the submitted work. Dr. Schwab reported receiving personal fees from ZimVie, Medtronic, and Stryker, consulting fees from Mainstay Medical, royalties from MSD, being a noncompensated shareholder in VFT Solutions and SeaSpine, and being a noncompensated executive board member of the ISSG, outside the submitted work. Dr. Shaffrey reported receiving personal fees from Medtronic, NuVasive/Globus, SI-BONE, and Proprio, outside the submitted work. Dr. Ames reported receiving royalties from DePuy Synthes, Zimmer Biomet Spine, K2M, Medicrea, Next Orthosurgical, NuVasive, and Stryker; consulting fees from DePuy Synthes, Medtronic, Medicrea, K2M, Agada Medical, and Carlsmed; research fees from Titan Spine, DePuy Synthes, and the ISSG; grants from the SRS; being on the editorial board of Operative Neurosurgery and Neurospine; being on the executive committee of the ISSG; being chair of the SRS Safety and Value Committee; and being the director of Global Spinal Analytics, outside the submitted work. This work was supported by the ISSG Foundation which receives funding support from DePuy Synthes, K2M, NuVasive, Orthofix, and Zimmer Biomet.