Allostatic Load in Cancer: A Systematic Review and Mini Meta-Analysis

Abstract

Background:

Individuals with cancer experience stress throughout the cancer trajectory. Allostatic load (AL), a cumulative multi-system measure, may have a greater value in stress assessment and the associated biological burden than individual biomarkers. A better understanding of the use of AL and its operationalization in cancer could aid in early detection and prevention or alleviation of AL in this population.

Purpose:

To consolidate findings on the operationalization, antecedents, and outcomes of AL in cancer.

Methods:

Seven databases (CINAHL, Ovid MEDLINE, Web of Science, APA PsycInfo, Scopus, Embase, and Cochrane CENTRAL) were searched for articles published through April 2020. The NIH tools were used to assess study quality.

Results:

Twelve studies met inclusion criteria for this review. Although variability existed in the estimation of AL, biomarkers of cardiovascular, metabolic, and immune systems were mostly used. Associations of AL with cancer-specific variables were examined mostly utilizing population-databases. Significant associations of AL with variables such as cancer-related stress, positive cancer history, post traumatic growth, resilience, tumor pathology, and cancer-specific mortality were found. Mini meta-analysis found that a one-unit increase in AL was associated with a 9% increased risk of cancer-specific mortality.

Conclusion:

This review reveals heterogeneity in operationalization of AL in cancer research and lack of clarity regarding causal direction between AL and cancer. Nevertheless, AL holds a significant promise in cancer research and practice. AL could be included as a screening tool for high-risk individuals or a health outcome in cancer. Optimal standardized approaches to measure AL would improve its clinical utility.

The diagnosis and symptom experience of cancer are highly stressful and potentially traumatic (Cordova et al., 2017). Posttraumatic stress disorder is common in cancer with a 22% incidence at 6 months post cancer diagnosis and a 6% incidence at 4 years post diagnosis (Chan et al., 2018). Stress-induced Takotsubo cardiomyopathy is reportedly found in approximately 10% of patients with cancer (Giza et al., 2017). The cancer treatment modalities also exacerbate stress levels. For instance, surgery-induced stress in cancer is known to have a systemic effect, involving inflammation, ischemic-reperfusion injury, sympathetic nervous system activation, and increased cytokine release (Chen et al., 2019). Negative psychological states are highly prevalent in patients with cancer (Kuhnt et al., 2016), which in turn, have been associated with lower white blood cell activity, reduced numbers of antibodies, and increased stress hormone response (Fancourt et al., 2016).

A healthy stress response is produced by a coordinated network of bidirectional feedback signals between the sympathetic nervous system (SNS), the hypothalamic-pituitary-adrenal (HPA) axis, and the immune system (Glaser & Kiecolt-Glaser, 2005; Smith & Vale, 2006; Sturmberg et al., 2017). This dynamic and adaptive regulatory process that maintains physiological stability during exposure to stressors is known as allostasis and was first introduced in 1988 (Sterling & Eyer, 1988). Allostasis is protective in the short-term, but long-term activation of the stress response systems results in allostatic load (AL). Allostatic load, first described by McEwen and Stellar (1993), refers to the cumulative biological burden exacted on the body’s systems due to repeated adaptation to stressors over time (McEwen, 2006). In the first stage of stress mediation, the acute stress response activates the primary mediators of AL including stress hormones (epinephrine, norepinephrine, cortisol) and antagonists (e.g., dehydroepiandosterone), in conjunction with pro- and anti- inflammatory cytokines (e.g., interleukin-6, tumor necrosis factor-alpha). The interaction of these primary mediators can result in primary allostatic effects such as anxiety, reduced sleep quality, and mood changes (McEwen, 2006; McEwen & Wingfield, 2003). In the second stage, a more long-term stress response—due to constant secretion of these primary mediators—results in secondary outcomes such as sub-clinical disturbances in cardiovascular, metabolic, and immune parameters (Juster et al., 2010). Finally, chronic stress dysregulation leads to the last stage of AL progression, which is known as the allostatic overload. In this final stage, the culmination of physiological dysregulations leads to tertiary or disease outcomes such as cardiovascular diseases, depression, cognitive decline, fatigue, cancer, or cellular aging, and eventually death (Juster et al., 2010; Leahy & Crews, 2012; McEwen & Wingfield, 2003).

The AL model proposes that the tertiary outcomes of AL can be predicted from extreme values of secondary outcomes and primary mediators (McEwen, 2000). Thus, health-related effects of stress can be quantified using a cumulative multisystem measure, termed as the allostatic load index (ALI), which may have greater value in stress assessment and the associated biological burden than individual biomarkers (McEwen, 2007; Seeman et al., 2001). The first study to operationalize AL and provide preliminary evidence of its predictive validity, used 10 biomarkers (Seeman et al., 1997). These biomarkers included 4 primary mediators—dehydroepiandrosterone sulfate (DHEA-S), urinary epinephrine, norepinephrine, and cortisol; and 6 secondary outcomes—systolic blood pressure (SBP), diastolic blood pressure (DBP), waist-hip ratio (WHR), high-density lipoprotein (HDL), total cholesterol (TC), and glycosylated hemoglobin (HbA1c). The ALI ranged from 0 to 10, and higher values of ALI indicated higher physiological strain (Seeman et al., 1997). This original operationalization of AL was further supported through the MacArthur Study of Successful Aging, which demonstrated that AL was a better predictor of mortality and decline in physical functioning than metabolic syndrome (Seeman et al., 2001). The original method has since formed the foundation for studies using AL. The construct validity of AL is established and studies have demonstrated common variance and statistical coherence between prominent primary mediators of stress response and secondary mediators reflecting biological alterations in autonomic, metabolic, and immune domains (Galen Buckwalter et al., 2016; Wiley et al., 2016). Although researchers agree that the AL construct is valid and its measurement should include biomarkers from neuroendocrine and immunological domains (Acheampong et al., 2020; Goldman et al., 2006; Karlamangla et al., 2002; Seeman et al., 2004), there is still no “gold standard” for AL estimation and disagreements exist on which combination of biomarkers best reflect a “gold standard” (Beckie, 2012; Duong et al., 2017; Gallo et al., 2014; Mauss et al., 2015). An alternative method to measure AL in clinical practice—the clinimetric criteria, was introduced in 2010 (Fava et al., 2010) and later revised in 2017 (Fava et al., 2017). The clinimetric criteria include the presence of an identifiable source of distress, along with one or more of the following manifestations occurring within 6 months after the onset of the stressor: psychiatric symptoms, psychosomatic symptoms, significant impairment in social or occupational functioning, and significant impairment in psychological well-being (Fava et al., 2010).

AL has been extensively studied among the elderly (Gruenewald et al., 2009; Karlamangla et al., 2002; Seeman et al., 1997; Zsoldos et al., 2018), and in different illness states such as psychosis (Piotrowski et al., 2020; Piotrowski et al., 2019), depression (Scheuer et al., 2018), cardiovascular disorders (Mazgelytė et al., 2019), anxiety disorder (Soria et al., 2013), and dementia (Cadar & Steptoe, 2019). However, the extent to which AL is examined in cancer is uncertain and to the best of our knowledge, no systematic review has synthesized the literature on measurement of AL in this population. Moreover, the extent of methodological variability in operationalization of AL in cancer is unclear, and it is not known how studies among cancer population has examined AL—whether as a predictor variable leading to cancer and related outcomes or as an outcome variable occurring in response to cancer and related precursors. Therefore, the purpose of this systematic review was to examine literature on use of AL in cancer. The specific review questions were:

1. How has AL been operationalized in studies related to cancer?

2. What cancer-related antecedents and outcomes of AL were examined?

3. What relationships among AL and cancer-related antecedents and outcomes were observed in this population?

For the purpose of this article, operationalization of AL refers to the domains or stress response systems included in estimating AL, biomarkers used within each domain, and method of composite AL calculation. Antecedents refer to variables that lead to or have an influence on AL and outcomes refer to variables which result from or are influenced by AL.

Method

This systematic review was guided by the Cochrane guidelines for systematic review (Higgins et al., 2019; Lefebvre et al., 2019; McKenzie et al., 2019), Garrard’s structured review method (Garrard, 2016), and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009). Keywords and search strings were identified in selected databases in an iterative process, and study titles were sampled for relevance through a pilot scoping search. The iterative process revealed that the Multi-Systemic Biological Risk was used as a proxy for allostatic load. In view of this finding, and as recommended by the Cochrane guidelines and Bates (Bates, 1989), a high-recall search using multiple strategies was done to retrieve articles which otherwise could have been missed.

Screening and Study Selection

Two authors (AM and AZD) determined the study selection criteria. These criteria were decided based on the focus of the systematic review—studies that have estimated AL in the context of cancer. A two-step screening process was adopted for the review. First, the article titles and abstracts were screened for the following inclusion criteria: (a) studies among individuals with cancer referring to AL, and (b) studies among general population examining associations of AL with cancer-specific variables. Those that met the inclusion criteria were deemed eligible for the full-text screening process. Studies were excluded based on the exclusion criteria: (a) ineligible populations—caregivers of cancer survivors, (b) studies not available in English, and (c) studies which referred to AL but measured biomarkers representing one or two systems without estimating a composite AL score.

Search Strategies and Data Sources

The investigators searched seven databases, including the Cumulative Index to Nursing and Allied Health Literature (CINAHL) Plus with Full Text, Ovid MEDLINE, Web of Science, APA PsycInfo, Scopus, Embase, and Cochrane Central Register of Controlled Trials (CENTRAL), during the period from 20 April 2020 through 23 April 2020. Search terms included allostatic load index OR allostatic load OR allostasis OR multi systemic biological risk, AND cancer OR neoplasm. The search was customized to each database. For instance, subject headings for CINAHL, mapping options for Embase, and thesaurus search for PsycInfo were used. Database-specific field designators and nesting features were used to improve search sensitivity and comprehensiveness. Additional information on the search strings are given in Supplemental table (Table S). The search was conducted with no date limits and included articles of all study designs and all languages. Ancestry and forward searching were performed on all articles that met the inclusion criteria.

Data Extraction and Synthesis

Garrard’s matrix method (Garrard, 2016) guided the data extraction and synthesis. Using a spreadsheet, data from the included articles were extracted in ascending chronological order with seven column topics: journal and author details; study design and purpose; study population; domains included; biomarkers within each domain; method of estimation; and study findings. Because the included studies were heterogeneous in their purposes and effect sizes, a meta-analysis of all the included studies could not be done. However, in keeping with the possibility of performing meta-analysis with small numbers of studies (Bragazzi, 2015; Goh et al., 2016), standard meta-analytic methods were used to calculate an overall effect of AL on cancer-specific mortality (n = 4) that reported hazard ratios. Meta-analyses were performed using the random effects model, calculating both Q-statistics and I² as indicators of heterogeneity, using STATA 16.0 (StataCorp, College Station, Texas, USA).

Quality Assessment

AM and HL assessed study quality using the design-specific National Institutes of Health (NIH) quality assessment tools (National Heart, Lung, and Blood Institute, 2019), with each tool consisting of 14 items (for cross-sectional, observational cohort, and intervention studies). Each study was assigned a score based on the items and was interpreted as good (10-14), fair (5-9), or poor (0-4), based on previous use of the tools by other systematic review authors (Kim et al., 2019) and in consensus with the authors of this review. The two quality assessors scored the studies independently and described the study characteristics that supported their judgment for methodological quality; these descriptions were used for discussion when there were disagreements.

Results

Search Results

The systematic search identified 448 articles, of which 22 were eligible for full-text screening. After full-text screening, 11 articles were included for final synthesis. Ancestry searching resulted in one additional article. The final number of articles was 12 (see PRISMA flowchart Figure 1).

Figure 1.

PRISMA flow chart illustrating article selection process.

Study and Participant Characteristics

The 12 articles selected for review were published between 2013 and 2020, from studies conducted in USA (Acheampong et al., 2020; Beydoun et al., 2019; Hughes Halbert et al., 2020; Levine & Crimmins, 2014; Mattei et al., 2010; Parente et al., 2013; Santacroce & Crandell, 2014; Xing et al., 2020), Taiwan (Hwang et al., 2014), Italy (Ruini et al., 2015), China (Ye et al., 2017), and Scotland (Robertson et al., 2017). As shown in Table 1, 6 studies used a cohort design (Acheampong et al., 2020; Beydoun et al., 2019; Hwang et al., 2014; Levine & Crimmins, 2014; Robertson et al., 2017; Xing et al., 2020), while 5 used a cross-sectional design (Hughes Halbert et al., 2020; Mattei et al., 2010; Parente et al., 2013; Ruini et al., 2015; Santacroce & Crandell, 2014). Only one study was a randomized controlled trial (Ye et al., 2017). Across the 12 studies, allostatic load was examined in the context of cancer in either of the two ways: (1) as a “predictor” variable associated with outcomes such as self-reported diagnosis of cancer (Mattei et al., 2010), cancer-specific mortality (Acheampong et al., 2020; Beydoun et al., 2019; Hwang et al., 2014; Levine & Crimmins, 2014; Robertson et al., 2017), breast cancer clinicopathology (Xing et al., 2020), and post-traumatic growth (Ruini et al., 2015), or (2) as an “outcome” variable associated with cancer-related stress (Santacroce & Crandell, 2014), history of breast cancer (Parente et al., 2013), resilience (Hughes Halbert et al., 2020), and a mentor-based supportive-expressive intervention program (Ye et al., 2017). These cancer-specific associations with AL were examined mostly among general population, with only 5 studies reporting results among individuals at risk for or with a diagnosis of cancer (Hughes Halbert et al., 2020; Ruini et al., 2015; Santacroce & Crandell, 2014; Xing et al., 2020; Ye et al., 2017).

Table 1.

Study Characteristics and Operationalization of Allostatic Load in Cancer (n = 12).

First author (year), country	Study design & Purpose	Study quality	Sample size, Gender %	Biomarkers comprising original Allostatic Load Index (Seeman et al., 1997)										Additional biomarkers used	Method of estimating AL score
				DHEA-S	Cortisol	Epinephrine	Norepinephrine	SBP	DBP	WHR	HDL	TC	HbA1C
Mattei et al. (2010), USA	Cross-sectional: Determine the association of AL to six chronic diseases including cancer in Puerto Rican older adults, and to compare strength of this association with metabolic syndrome	Fair	1116; 72% female	✓	✓	✓	✓	✓	✓		✓	✓	✓	WC, use of testosterone or medications for HTN, DM, or hypercholesterolemia	Participants categorized as high risk on biomarkers using clinical cut-off values; sum of parameters coded as high risk obtained as AL score; categories of AL created based on number of parameters above or below the cut-off as 0–2, 3–5, ≥ 6
Parente et al. (2013), USA	Cross-sectional: To compare AL scores of women with history of BC to those without, stratified by race and to evaluate whether the interaction between race and BC predicts elevated AL levels	Fair	4875; 100% female					✓	✓		✓	✓	✓	Heart rate, BMI, CRP, albumin	Biomarkers dichotomized as 0 & 1 using clinical cut-off values; sum of parameters coded 1 obtained as AL score; converted AL score into a dichotomous variable, with high or elevated AL defined as ≥ 3
Santacroce and Crandell (2014), USA	Cross-sectional: To assess the feasibility of studying AL and describe preliminary findings concerning psychological distress, PTSS/PTSD, health behavior and AL in AYA childhood cancer survivors and their siblings	Fair	Survivors - 8, 25% female; Siblings - 8, 50% female		✓			✓	✓		✓	✓	✓	Cortisol morning response & diurnal slope, salivary α- amylase (morning response & diurnal slope), diurnal slope, heart rate, BMI, glucose, triglycerides, insulin, hs-CRP, IL-6	Biomarkers dichotomized as 0 & 1 using population-specific high risk quartiles; sum of parameters coded “1 = high risk” obtained as AL composite score; basic AL computed using 12 biomarkers and standard AL computed using 16 biomarkers; The AL composite was considered in the high range if the participant scored 1 on at least 25% of the biomarkers in the composite; high AL defined as ≥ 3 (basic AL) and ≥ 4 (standard AL)
Levine and Crimmins (2014), USA	Retrospective cohort: To compare how well AL, the FRS and biological age predict subsequent 10-year all-cause and disease-specific mortality	Good	9942; 52.9% female					✓	✓	✓	✓	✓	✓	Pulse rate, albumin, CRP For expanded AL: creatinine, CMVOD, alkaline phosphatise, FEV1, urea nitrogen	Participants categorized as high risk on 9 biomarkers using clinical cut-off values; sum of parameters coded as high risk obtained as normal AL score, expanded AL calculated using high risk quintiles of additional 5 biomarkers, continuous AL calculated using two-tailed z-scores for the 14 biomarkers
Hwang et al. (2014), Taiwan	Prospective cohort: To evaluate the role of AL, either static or dynamic measurements, in predicting 10-year cause-specific mortality and all-cause mortality in Taiwanese older adults	Good	Baseline AL - 1023, 57.67% male; Dynamic AL - 757, 56.27% male	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	IGF-1, dopamine, triglycerides, fasting glucose, BMI, WBC, neutrophils, IL-6, albumin, creatinine	Biomarkers considered positive if they fell below the 10th and above the 90th percentiles; count of biomarkers falling in high risk deciles obtained as AL score; participants divided into quintiles based on their AL scores—Q1 (AL score of 0 -1); Q2(AL score of 2); Q3(AL score of 3); Q4(AL score of 4-5); and Q5(AL score of 6-12); dynamic AL included stratification of participants based on AL score changes from 2000 to 2006—no change, decline, slow increase, fast increase
Ruini et al. (2015), Italy	Cross-sectional: To evaluate the relationship between PTG and AO among BC survivors and healthy stressed women matched on age and sociodemographic variables	Fair	Survivors - 60, Healthy women - 60; Both groups 100% female											Presence of chronic stress, presence of psychiatric symptoms, low psychological well-being, presence of psychiatric symptoms combined with low psychological well-being	Used clinimetric criteria for categorizing as AO (a) scoring positive on specific PSI items (b) scoring higher than 75th percentile on at least two SQ distress scales (c) scoring lower than 25th percentile on at least three PWB scales (or two if one of the scales is environmental mastery) and (d) scoring higher than 75th percentile on one SQ scale and contemporaneously lower than 25th percentile on two PWB scales
Ye et al. (2017), China	RCT: To assess effect of BRBC program on 3 and 5-year cancer-specific survival, anxiety, QOL, depression, resilience, and ALI in women with BC	Good	226; 100% female		✓			✓	✓	✓				BMI, resting pulse, SDRR, RBC, WBC, Hb, serotonin, CRP, IL-6, CD4+/CD8+	Participants received 1 point for each biomarker if it was abnormal; sum of points obtained as AL score; identified ALI trajectories as chronic, delayed, recovery, and resilient
Robertson et al. (2017), Scotland	Retrospective cohort: To assess how well AL predicts 5-year and 10-year all-cause and cause-specific mortality	Good	4,488; 56.2% female					✓	✓	✓	✓	✓	✓	Pulse rate, CRP, adjusted for use of medications	Biomarkers dichotomized as 0 & 1 using population-specific high risk quartiles; sum of parameters coded “1 = high risk” obtained as AL score
Beydoun et al. (2019), USA	Retrospective cohort: To examine the mediating-moderating effect of the AL score on the association between the DASH dietary pattern and all-cause, CVD and cancer-specific mortality risks	Good	11,630; 51.5% female					✓	✓		✓	✓	✓	BMI, TC/HDL ratio, CRP, albumin, creatinine clearance	Participants categorized as 0, 0.5, or 1 on each biomarker using risk categories; sum of unweighted biomarker scores obtained as AL score
Xing et al. (2020), USA	Retrospective cohort: To examine the association of pre-diagnostic AL with unfavorable tumor clinicopathologic features such as invasive tumor behavior, higher tumor grade, larger tumor size, and ER negative status in Black BC survivors	Good	AL measure 1 - 229, AL measure 2 - 409; 100% female					✓ ✓	✓ ✓		✓	✓		AL measure 1: triglycerides, glucose, WC, consideration of LDL if TC ≤ 240 mg/dL, use of medications for HTN, DM, or hypercholesterolemia AL measure 2: BMI, glucose, WC, albumin, eGFR, use of diabetes, hypertension, or hypercholesterolemia medication	AL estimated using 2 computation methods—AL measure 1 (lipid profile-based) and AL measure 2 (inflammatory profile-based); biomarkers dichotomized as 0 & 1 using cut-off values; sum of respective parameters coded 1 obtained as AL measures 1 and 2; median AL score for both measures used as cut-off to dichotomize each measure—Lower AL, 0–3 points, Higher AL, 4–8 points. Cohen’s Kappa statistic (k) calculated to test the agreement between AL measures 1 and 2
Acheampong et al. (2020), USA	Retrospective cohort: To examine the association between MSBR index and cancer mortality	Good	13,628; 46.6% to 59.7% female (range in MSBR quartiles)					✓	✓					Pulse rate, HOMA-IR, triglycerides, waist circumference, WBC, CRP	Participants received scores of 0, 1, or 2 on each biomarker based on clinical cut-off values or empirical evidence; sum of scores obtained as MSBR index; ranked the total index into quartiles
Halbert et al. (2020), USA	Cross sectional: To examine relationships between two aspects of resiliency (ability to adapt and bounce back), AL, and sociodemographic factors among men undergoing prostate biopsy	Fair	47; 100% male	✓				✓	✓	✓		✓	✓	HR, BMI, CRP, albumin, creatinine	Biomarkers dichotomized as 0 & 1 using population-specific high risk quartiles; sum of parameters coded “1 = high risk” obtained as AL score

Note. Check marks indicate biomarkers of the original AL index used; blank columns indicate not used in respective studies. ALI = Allostatic Load Index; AL = Allostatic Load; AO = Allostatic Overload; AYA = Adolescent Young Adult; BC = Breast Cancer; BMI = Body Mass Index; DBP = Diastolic Blood Pressure; BRBC program = “Be Resilient to Breast Cancer” program; CD4⁺/CD8⁺ = Cluster of Differentiation 4/Cluster of Differentiation 8; CMVOD = Cytomegalovirus Optical Density; CRP = C-Reactive Protein; CVD = Cardiovascular Disease; DASH = Dietary Approaches to Stop Hypertension; DHEA-S = Dihydroepiandrosterone Sulphate; DM = Diabetes Mellitus; eGFR = estimated Glomerular Filtration Rate; ER = Estrogen Receptor; FEV1 = Forced Expiratory Volume at 1 s; FRS = Framingham Risk Score; Hb = Hemoglobin; HbA1c = Glycosylated Hemoglobin; HDL = High Density Lipoprotein; HOMA-_IR = Homeostasis Model Assessment for Insulin Resistance; hs-CRP = high-sensitivity C-reactive Protein; HTN = Hypertension; IGF-1 = Insulin Growth Factor-1; IL-6 = Interleukin-6; LDL = Low Density Lipoprotein; MSBR = Multi Systemic Biological Risk; PSI = Psycho Social Index; PTG = Post Traumatic Growth; PTSS/PTSD = Post Traumatic Stress Symptoms/Post Traumatic Stress Disorder; PWB Scale = Psychological Well-Being Scale; QOL = Quality of Life; RBC = Red Blood Cells; RCT = Randomized Controlled Trial; SBP = Systolic Blood Pressure; SEBAS = Social Environment and Biomarkers of Aging; SQ = Symptom Questionnaire; SDRR = Standard Deviation of R-R intervals; TC = Total Count; WBC = White Blood Cells; WC = Waist Circumference; WHR = Waist-hip Ratio.

Table 2 describes the details of the study populations. Eight studies used data from large population databases (Acheampong et al., 2020; Beydoun et al., 2019; Hwang et al., 2014; Levine & Crimmins, 2014; Mattei et al., 2010; Parente et al., 2013; Robertson et al., 2017; Xing et al., 2020). One such database was the National Health and Nutrition Examination Survey (NHANES)—a nationally representative, cross-sectional survey of the non-institutionalized US population, which was used in 4 studies. Other databases were from the Boston Puerto Rican Health Study, the Scottish Health Survey, the Women’s Circle of Health Follow-Up Study, and the Social Environment and Biomarkers of Aging cohort study.

Table 2.

Study Population and Findings Pertaining to Allostatic Load and Cancer (n = 12).

First author (year)	Population	Mean AL (SD) ____________________________________ Confounders statistically adjusted for	Associations with increased (↑) AL and other study findings
Mattei et al. (2010)	Boston Puerto Rican Health Study dataset (years 2004-2008), Puerto Rican adults aged 45–75 years	3.8 (1.7)	AL not significantly associated with self-reported cancer [OR (95%CI): 0.43 (0.14–1.29)], p = 0.130, used ≥ 6 parameters]
		Age, sex, smoking, alcohol intake, physical activity, total dietary fat intake, total energy intake
Parente et al. (2013)	NHANES database (years 1999-2008), black and white women aged 35–85 years	NR	History of BC significant predictor of elevated AL in black women [OR (95%CI): 2.08 (1.02, 4.22), p ≤ 0.05], but not in white women [OR (95% CI): 0.94 (0.62, 1.42)]. Interaction between black and having a history of BC significant (β = 0.21, p ≤ 0.05) in predicting ↑ AL Black women with history of BC 16.4% points more likely to have ↑ AL than blacks without history of BC
		Age, income, education, insurance type, smoking status, alcohol intake, physical activity, CAD, cancer other than BC
Santacroce and Crandell (2014)	AYA childhood cancer survivors & their biological siblings, both aged 15–29 years	Survivors: Basic AL: 3.63 (1.30), Standard AL: 4.50 (1.41) Siblings: Basic AL: 2.63 (2.00), Standard AL: 3.12 (1.95)	Cancer survivors had higher AL (basic and standard) than siblings Feasibility: excellent adherence to salivary sample collection (93.75%); 97.2% of the survivors’ samples had sufficient volume for the assays
		Predictors / associations of AL not studied
Levine and Crimmins (2014)	NHANES III database (years 1988-1994), adults aged >30 years	Normal AL: 2.12 (1.47) Expanded AL: 3.20 (2.08) Continuous AL: 9.25 (3.36)	For the full age range (30+), normal AL predicted 10-year all-cause mortality with 67.9% accuracy (AUC = 0.6791), expanded AL with 77% accuracy, and continuous AL with 81.5% accuracy as against biological age with 87.5% accuracy (all Bonferroni p < 0.0001). Among participants ages 50 to 69, continuous Allostatic Load predicted mortality with about 72% accuracy, which was significantly better than other AL measures. Participants with the highest scores of normal AL had 2 times increased risk of cancer-specific mortality compared to those with the lowest scores, risk increased to 2.1 with use of expanded AL and 3.7 with continuous AL; biological age had the strongest association with cancer-specific mortality (13.5 times)
		Chronological age, sex
Hwang et al. (2014)	SEBAS dataset (year 2000), adults aged >54 years	Baseline AL: 4.02 (2.20)	Each point increase in baseline AL associated with 18% incremental risk of 10-year cancer-specific mortality [HR (95%CI) = 1.18 (1.08–1.29), p = 0.001] Higher AL score quintile significantly associated with higher 10-year cancer-specific mortality (p = 0.008) Association of dynamic AL with cancer-specific mortality not reported
		Age, sex
Ruini et al. (2015)	BC survivors and healthy women with major life stressors	NR	51.7% of BC survivors presented with AO, BC survivors without AO had highest levels of PTG [M (SD) = 71.30 (4.34), p = 0.064] Among those with AO, BC survivors had significantly higher scores on PTG scales of personal strength [F (3, 106) = 3.816, p = 0.012] and spiritual changes [F (3, 106) = 3.091, p 9= 0.031]. Two groups without AO exhibited no significant differences on the PTG scales. BC survivors scored significantly higher than healthy stressed women with AO on the new possibilities (p = 0.036), personal strengths (p = 0.002), and spiritual changes (p = 0.050) scales.
		Age, time since event, marital status, work status, event’s objective negative impact
Ye et al. (2017)	Women with metastatic BC	Intervention group: 4.75 (1.43) Control group: 4.47 (1.61)	Significant reduction in AL in intervention group as compared to control group at 6 month post intervention (ES = 0.75, p = 0.0009) and 12 month post intervention (ES = 0.90, p < 0.0001)
		Influencing factors not specified
Robertson et al. (2017)	Scottish Health Survey dataset (year 2003), individuals aged ≥ 16 years	NR	AL not associated with risk of cancer-specific mortality at 10 year [HR (95% CI) = 1.04 (0.92–1.18), p = 0.534]
		Unadjusted for cause-specific mortality
Beydoun et al. (2019)	NHANES database (years 2001-2010), adults aged ≥ 30 years	2.82 (SEM: 0.02)	ALI not associated with cancer-specific mortality rate [Loge(HR), 95% CI = 0.065 (– 0.083, 0.21)] No statistically significant interaction effects between DASH and ALI scores in relation to cancer-specific mortality, suggesting the absence of moderation by ALI
		Age, sex, race, education, marital status, poverty income ratio, smoking status, alcohol intake, physical activity, self-rated health, weight status
Xing et al. (2020)	WCHSF database (years 2014-2018), Black non metastatic BC survivors aged 20–75 years	AL measure 1: 3.09 (1.46) AL measure 2: 3.15 (1.61)	Moderate agreement between prediagnostic AL measures 1 and 2 (k = 0.504) Higher AL measure 1 associated with increased odds of poorly differentiated tumor [OR (95% CI) = 2.16 (1.18–3.94), p = 0.01] Higher AL measure 2 associated with increased odds of poorly differentiated tumor [OR (95% CI) = 1.60 (1.02–2.51), p = 0.04], and larger tumor size [OR (95% CI) = 1.58 (1.01–2.46), p = 0.04]
		Age at diagnosis, birthplace, marital status, menopausal status, family history of BC
Acheampong et al. (2020)	NHANES III database (years 1988-1994), adults aged ≥ 20 years	NR	Highest quartile of MSBR index associated with 64% increased risk for cancer-specific mortality [HR (95%CI) = 1.64 (1.13–2.40), p = .02], while a continuous MSBR index associated with 7% increased risk [HR (95%CI) = 1.07 (1.01–1.14), p = .02]. Association of MSBR with cancer-specific mortality differed by BMI (p = 0.02) Positive association of MSBR with risk of cancer-specific mortality particularly for those with BMI ≥ 25 [HR (95%CI) = 1.12 (1.05–1.19), p = 0.001], as against those with BMI < 25 [HR (95%CI) = 1.04 (0.92–1.18), p = 0.53]
		Age, sex, ethnicity, education, fasting status, health insurance, urbanization, current tobacco use, alcohol use, medication, HEI, physical activity, BMI
Halbert et al. (2020)	Minority and non-minority male Veterans at risk for prostate cancer	2.71 (1.5)	AL significantly higher in men who reported ability to bounce back [Mean (SD) = 3.3 (1.3)] than those who reported less resilience [Mean (SD) = 2.3 (1.6)], (t = -2.36, p = 0.02). AL higher (non-significant) in men who reported ability to adapt [Mean (SD) = 3.0 (1.6)] than those who had less resilience [Mean (SD) = 2.2 (1.3)], (t = -1.59, p = 0.12)
		Only bivariate analyses conducted

Note. AL = Allostatic Load; AO = Allostatic Overload; AUC = Area under Curve; AYA = Adolescent Young Adult; BC = Breast Cancer; BMI = Body Mass Index; CAD = Coronary Artery Disease; CI = Confidence Interval; HEI = Healthy Eating Index; HR = Hazard Ratio; M = Mean; MetS = Metabolic syndrome; MSBR = Multi Systemic Biological Risk; NHANES = National Health and Nutrition Examination Survey; NR = Not Reported; OR = Odds Ratio; PTG = Post Traumatic Growth; SD = Standard Deviation; SEBAS = Social Environment and Biomarkers of Aging; SEM = Standard Error of the Mean; WCHSF = Women’s Circle of Health Follow-Up.

A total of 47,520 participants were enrolled in these 12 studies. Among the studies using population databases, sample sizes ranged from 409 (Xing et al., 2020) to 13,628 (Acheampong et al., 2020), with an average of about 5,889 participants. Across the studies, female participants were predominant, with 4 studies having only female participants (Parente et al., 2013; Ruini et al., 2015; Xing et al., 2020; Ye et al., 2017) and the others with percentage female ranging from 25% (Santacroce & Crandell, 2014) to 72% (Mattei et al., 2010). Only one study included exclusively men, as it examined the association of AL and resiliency in men undergoing prostate biopsy (Hughes Halbert et al., 2020). Study characteristics and population are shown in Tables 1 and 2.

Study Quality

Seven of the 12 studies received a good quality rating (See Table 1). Overall, all the 12 studies had clearly specified study objectives and populations and used valid and reliable measures to assess AL and other variables of interest. However, few aspects of methodological quality were widely ignored; for example, the exposure of interest was assessed more than once only in one study (Hwang et al., 2014).

Operationalization of AL in Cancer

Tables 1 and 3 describe how AL was operationalized in the 12 studies. Overall, studies varied in the choice of systems involved in stress response for AL estimation. The cluster of cardiovascular-metabolic-immune systems was used to estimate AL in 5 studies (Acheampong et al., 2020; Beydoun et al., 2019; Levine & Crimmins, 2014; Parente et al., 2013; Robertson et al., 2017); of which, one utilized this cluster to estimate Multi-Systemic Biological Risk (MSBR), a proxy for AL (Acheampong et al., 2020). The cluster of neuroendocrine-cardiovascular-metabolic-immune systems was used in 4 studies (Hughes Halbert et al., 2020; Hwang et al., 2014; Santacroce & Crandell, 2014; Ye et al., 2017). One study included the cluster of neuroendocrine-cardiovascular-metabolic systems in AL estimation (Mattei et al., 2010), while another study (Ruini et al., 2015) used the clinimetric criteria. All the studies using physiological biomarkers obtained samples from blood, saliva, or urine. Two studies reported estimation of more than one AL measure using different clusters of domains. One study computed AL measure 1 involving cardiovascular-metabolic systems and AL measure 2 involving cardiovascular-metabolic-immune systems (Xing et al., 2020). The other study computed normal AL using the cluster of cardiovascular-metabolic-immune systems; expanded AL using the cluster of cardiovascular-metabolic-immune-renal-hepatic-respiratory systems; and continuous AL using continuous z-scores for the same system cluster as expanded AL (Levine & Crimmins, 2014). On the whole, the majority of the studies used cardiovascular, metabolic, and immune system measures. All the studies that did not use population databases examined biomarkers from these 3 domains. Neuroendocrine measures were used only in 5 studies and only 2 population databases (Hwang et al., 2014; Mattei et al., 2010) had neuroendocrine assessments available (see Table 1).

Table 3.

Summary of Biomarkers used in Estimating Allostatic Load Index in Cancer (n = 12).

Group	Domain	Biomarkers used (units)	Description	Reported high risk cut off values or risk categories with scoring
Primary mediators	Neuroendocrine	Salivary cortisol morning response (log-μg/dL) and diurnal slope (log-μg/d/hr)	Adrenal glucocorticoid and indicator of HPA-axis activity	Cut-off not reported
		12-hour urine cortisol (μg/g creatinine)		≥41.5 (males), ≥49.5 (females) <0.08 mol/mol creatinine and >0.48 mol/mol creatinine
		Salivary α- amylase morning response (log-μg/dL) and diurnal slope (log-μg/d/hr)	Enzyme synthesized in pancreatic gland and salivary glands for enzymatic cleavage of glucose	Criterion cut-off not reported (population-specific quartiles used)
		12-hour urine epinephrine (μg/g creatinine)	Catecholamine, neurotransmitter and indicator of SNS activity	≥2.8 (males), ≥3.6 (females) >0.076 mol/mol creatinine
		12-hour urine norepinephrine (μg/g creatinine)	Catecholamine, neurotransmitter and indicator of SNS activity	≥30.5 (males), ≥46.9 (females) <0.108 mol/mol creatinine and > 0.343 mol/mol creatinine
		12-hour urine dopamine (mol/mol creatinine)	Catecholamine, neurotransmitter and indicator of SNS activity	<0.867 and >2.25
		DHEA-S (ng/mL)	Adrenal hormone and functional HPA-axis antagonist	≤589.5 (males), ≤368.5 (females) <20 μg/dL
		Serotonin	Neurotransmitter, regulates mood and social behavior, appetite and digestion, sleep, memory, and sexual desire and function	Not reported
		Insulin-like growth factor1 (nmol/L)	Growth hormone and regulates cellular DNA synthesis	<50 and >170
	Cardiovascular	SBP (mmHg)	Indicator of intravascular pressure at end of left ventricular contraction	≥140 <114 and >166 Categories: ≥150 = score of 1, 120 to <150 = score of 0.5, <120 = score of 0 Categories: ≥140 or use of anti-hypertensives = score of 2, 120 to 139 = score of 1, <120 = score of 0
		DBP (mmHg)	Indicator of intravascular pressure at end of left ventricular relaxation	≥90 <70 and >97 Categories: ≥90 = score of 1, 80 to <90 = score of 0.5, <80 = score of 0 Categories: ≥90 or use of anti-hypertensives = score of 2, 80 to 89 = score of 1, <80 = score of 0
		Pulse (beats/min)	Heart rate	≥90 Categories: ≥ 100 = score of 2, 61 to 99 = score of 1, ≤ 60 = score of 0
		SDRR	Physiological phenomenon of variation in the time interval between heartbeats measured by the variation in beat-to-beat interval	Cut-off not reported
		Red blood cells	Indicator of tissue oxygenation	Cut-off not reported
		Hemoglobin	Protein in red blood cells that transports oxygen and carbon dioxide between lungs and body’s tissues	Cut-off not reported
	Immune	hs-CRP, CRP (mg/L)	Acute phase inflammatory protein	>3 Categories: ≥3 = score of 1, 1 to <3 = score of 0.5, <1 = score of 0 Categories: > 1 mg/dL = score of 2, >0.21 to 1 mg/dL = score of 1, ≤ 0.21 = score of 0
		Interleukin-6 (pg/mL)	Pro-inflammatory cytokine and anti-inflammatory myokine stimulating immune response	>6.5
		White blood cell count	Indicator of inflammation	<4.5 and >8 (×109/L) Categories: > 11,001 or < 1500 cells/mcL = score of 2, 4500 to 11,000 cells/mcL = score of 1, 1500 to 4500 cells/mcL = score of 0
		Neutrophils (%)	Type of white blood cells that lead the immune response	<44 and >70
		CD4+/CD8+	Ratio of T helper cells to cytotoxic T cells and indicator of immune function	2.0
		Albumin (g/dL)	Early indicator of subclinical renal damage	<4 <3.8 <4.2 mg/dL Categories: <3 = score of 1, 3 to <3.8 = score of 0.5, ≥3.8 = score of 0
		Estimated glomerular filtration rate (mL/min)	Flow rate of filtered fluid through the kidney and indicator of renal function	<59
		Creatinine (mg/dL)	Breakdown product of muscle metabolism and indicator of renal function	≥1.3 >1.4
		Creatinine clearance (mL/min/1.73m2)	Volume of blood plasma that is cleared of creatinine per unit time, measure of renal filtration function	Categories: <30 = score of 1, 30 to <60 = score of 0.5, ≥ 60 = score of 0
		Urea nitrogen	Indicator of renal function	≥18
		CMV optical density	Indicator of IgG antibodies to cytomegalovirus	≥3
		Alkaline phosphatase	Enzyme present in all body tissues	>101
		FEV1	Maximal amount of air forcefully exhaled in one second	≥39.33
Secondary outcomes	Metabolic	Total cholesterol (mg/dL)	Basic element of steroid hormones, indicator of atherosclerotic risk	≥240 ≤ 240 and LDL >130 <155 and >250 Categories: ≥240 = score of 1, 200 to <240 = score of 0.5, <200 = score of 0
		HDL (mg/dL)	Cardioprotective form of cholesterol, indicator of atherosclerotic risk	<50 <40 <33 Categories: <40 = score of 1, 40 to <60 = score of 0.5, ≥60 = score of 0
		Total cholesterol / HDL ratio	Indicator of atherosclerotic risk	Categories: ≥6 = score of 1, 5 to <6 = score of 0.5, <5 = score of 0
		Triglycerides (mg/dL)	Cardio-damaging form of fat, important source of energy	≥ 150 <55 and >205 Categories: > 200 = score of 2, 150 to 199 = score of 1, < 150 = score of 0
		BMI (kg/m2)	Indicator of obesity based on weight and height	≥30 <19.95 and >28.83 Categories: ≥30 = score of 1, 25 to <30 = score of 0.5, <25 = score of 0
		Waist-to-hip ratio	Indicator of location of adipose tissue deposits based on ratio of waist circumference to hip circumference	>0.90 (males), >0.85 (females) <0.8 and >1
		Waist circumference (cm)	Measurement around the abdomen at the level of the umbilicus	≥88 >102 (males), >88 (females) Categories: Males: > 102 = score of 2, 94–102 = score of 1, < 94 = score of 0 Females: > 88 = score of 2, 80–88 = score of 1, < 80 = score of 0
		Glucose (mg/dL)	Primary source of energy	≥ 110
		Fasting glucose (mg/mL)		≥126
		Insulin	Pancreatic hormone for regulating glucose levels
		HbA1c (%)	Average glucose level over the previous 12 weeks, indicating degree of blood glucose regulation	≥6.4 ≥7.0 <4.8 and >7.1 Categories: ≥6.5 = score of 1, 5.7% to <6.5% = score of 0.5, <5.7% = score of 0
		HOMA-IR	Measure of insulin resistance	Categories: > 4.65 or diabetes diagnosis = score of 2, ≥ 2.6 to 4.65 = score of 1, < 2.6 = score of 0
Tertiary outcomes	Use of medications	Anti-hypertensives, insulin, OHA, lipid-lowering or testosterone		OHA and HbA1c ≤7.0 Anti-hypertensives and SBP ≤140 and DBP ≤90 Lipid-lowering medications and HDL ≥40 and TC <240
	Self-reported diagnoses	Heart attack, heart disease, stroke, arthritis, cancer		NA
	Clinimetric criteria	Chronic stress, psychiatric symptoms, low psychological well-being		NA

Note. BMI = Body Mass Index; CD4⁺/CD8⁺ = Cluster of Differentiation 4/Cluster of Differentiation 8; CMVOD = Cytomegalovirus Optical Density; DBP = Diastolic Blood Pressure; DHEA-S = Dihydroepiandrosterone Sulphate; FEV = Forced Expiratory Volume; HbA1c = Glycosylated Hemoglobin; HDL = High Density Lipoprotein; HOMA-_IR = Homeostasis Model Assessment for Insulin Resistance; HPA-axis = Hypothalamic-Pituitary-Adrenal axis; hs-CRP = high-sensitivity C-reactive Protein; IL-6 = Interleukin-6; LDL = Low Density Lipoprotein; NA = Not applicable; OHA = Oral Hypoglycemic Agents; SBP = Systolic Blood Pressure; SDRR = Standard Deviation of R-R intervals; SNS = Sympathetic Nervous System.

Apart from differences in use of system-clusters, the number and choice of biomarkers within each domain also varied across studies (see Tables 1 and 3). For instance, the measures of cardiovascular system domain mainly included systolic blood pressure (SBP), diastolic blood pressure (DBP), and heart rate; but Ye et al. (Ye et al., 2017) included standard deviation of R-R intervals (a metric of heart rate variability) along with SBP and DBP in this domain. In the immune system domain, although C-reactive protein (CRP) or Interleukin-6 (IL-6) were generally included, studies also included albumin, white blood cell count, glomerular filtration rate, neutrophils, and creatinine. The total number of biomarkers used in computing ALI within a study ranged from 7 (Acheampong et al., 2020) to 20 (Hwang et al., 2014). Researchers commonly measured 4-6 biomarkers of the original AL index (Table 1), with only one study (Hwang et al., 2014) measuring all the 10 biomarkers from the original AL index (Seeman et al., 1997).

There were variations in how studies estimated the composite AL score. Except the study that used clinimetric criteria, all the other 11 studies assigned a score of 1 for each biomarker if the participant fell in the high-risk category for that biomarker. The details of risk-scoring and threshold values of the risk categories are described in Tables 1 and 3. Across the 11 studies, assigning high-risk scoring was primarily done in 2 ways: using clinical cut-off values or cut-off using population-specific quantiles. Six studies used cut-off values based on clinically or empirically significant threshold of risk for disease (Acheampong et al., 2020; Beydoun et al., 2019; Mattei et al., 2010; Parente et al., 2013; Xing et al., 2020; Ye et al., 2017). Four studies used population-specific quartiles/percentiles/deciles to assign high-risk scoring. In 3 studies, participants were deemed high risk on a specific biomarker if they scored in the poorest 25% of the sample on that biomarker (Hughes Halbert et al., 2020; Robertson et al., 2017; Santacroce & Crandell, 2014). For example, those at high risk on SBP were in the upper quartile of the sample and the high-risk quartile for HDL or DHEA-S was the lowest quartile. In the fourth study (Hwang et al., 2014), extreme values of both tails (< 10th and > 90th percentiles) for most of the biomarkers were used for considering high-risk categories in accordance with the method outlined by Seplaki et al. (2005). In one study, researchers used both ways of risk scoring—clinical cut-off values for normal AL measure and high-risk quantiles for expanded AL measure (Levine & Crimmins, 2014). Irrespective of differences in the methods used in assigning a high-risk score, all the studies computed the final AL score by summing up the scores assigned for each biomarker—a higher AL score signifying greater dysregulation (Parente et al., 2013). The range of composite AL score varied according to the number of biomarkers used. Finally, the AL composite was included in statistical analysis, either as a continuous measure or categorical measure (see Table 1). Six studies included the AL composite as categorical variable (Acheampong et al., 2020; Hwang et al., 2014; Mattei et al., 2010; Parente et al., 2013; Santacroce & Crandell, 2014; Xing et al., 2020) while the other 5 used it as a continuous variable. One study (Levine & Crimmins, 2014) specifically reported using a continuous, z-score measure of AL in accordance with the method outlined by Seplaki et al (Seplaki et al., 2005).

Use of medications to control hypertension, diabetes, or hypercholesterolemia were considered only in 4 studies; of which three considered medication use directly in AL scoring (Mattei et al., 2010; Robertson et al., 2017; Xing et al., 2020), while the fourth included medication use as a covariate associated with biomarker level (Acheampong et al., 2020). For instance, in one study, if the participant was on anti-hypertensive medications, the SBP and DBP were increased by 10 mmHg and 5 mmHg respectively and for beta-blockers, the HDL was increased by 10% (Robertson et al., 2017). The researchers made these adjustments based on empirical evidence of drug dosing and their effects on these parameters. In other instances, a score was assigned for use of medications (Mattei et al., 2010; Xing et al., 2020).

Relationship of AL With Cancer-Related Antecedents and Outcomes

Four studies examined cancer-related antecedents that impacted AL (see Table 2). One study reported that after adjusting for sociodemographic characteristics, history of breast cancer was significantly associated with elevated AL in Black women, but not in White women. This study used the NHANES database to examine association between breast cancer and allostatic load scores stratified by race in 4875 women. Black women constituted 24.9% of the total sample (n = 1214), of which 35 had a history of breast cancer. Also, an interaction between being Black and having a history of breast cancer was found to be significant in predicting elevated AL scores after adjusting for demographic, behavioral, and comorbidity characteristics (Parente et al., 2013). Another study reported that among Veterans at risk for adverse prostate cancer outcomes, men who had greater resilience to bounce back from injury, illness, or other hardships, had higher AL (Hughes Halbert et al., 2020). A third study focused on the impact of childhood cancer-related stress on AL and found that survivors of childhood cancer had higher AL composites than their siblings (Santacroce & Crandell, 2014). Finally, another antecedent examined was a psycho-social intervention (mentor-based, supportive expressive intervention program) for individuals with metastatic breast cancer, which led to significant reduction in AL at 6- and 12-months’ post intervention (Ye et al., 2017).

The remaining studies examined cancer-related outcomes associated with AL. Cancer-specific mortality was examined in 5 studies (Acheampong et al., 2020; Beydoun et al., 2019; Hwang et al., 2014; Levine & Crimmins, 2014; Robertson et al., 2017). Two studies did not find that AL predicted cancer-specific mortality (Beydoun et al., 2019; Robertson et al., 2017). The other 3 studies, however, reported significant association of AL and cancer-specific mortality, albeit, with different risk percentages. One study, which operationalized AL in 3 ways and compared different methods to assess mortality risk, reported that participants with the highest scores of normal AL had 2 times increased risk of cancer-specific mortality compared to those with the lowest scores, and this risk increased to 2.1 with use of expanded AL and 3.7 with continuous AL (Levine & Crimmins, 2014). However, among the various methods to assess mortality risk, biological age had the strongest association with cancer-specific mortality (see Table 2). Another study that examined the MSBR, a proxy for AL, reported that participants in the highest quartile of MSBR had 64% increased risk for cancer mortality compared to the lowest quartile (Acheampong et al., 2020). The increased risk was primarily driven by the immune domain with a stronger association observed in overweight/obese individuals (Acheampong et al., 2020). The fifth study observed an 18% incremental risk of 10-year cancer-specific mortality with each point increase in AL (Hwang et al., 2014).

The second outcome that was examined was posttraumatic growth (PTG), where the researchers compared breast cancer survivors with and without allostatic overload (AO) and healthy stressed women with and without AO (Ruini et al., 2015). Women with breast cancer had higher scores on PTG scales reflecting new possibilities, personal strengths, and spiritual changes than healthy women with AO (Ruini et al., 2015). Another study done among Black women with breast cancer reported that higher pre-diagnostic AL predicted higher tumor grade and larger tumor size (Xing et al., 2020). Finally, one study reported that increasing categories of AL score were not significantly associated with self-reported cancer among Puerto Rican older adults (Mattei et al., 2010).

Meta-Analysis

The mini meta-analysis included 30,769 subjects from four studies that estimated hazard ratios (HR) of cancer mortality risk according to AL score. The overall effect of AL and cancer-specific mortality was calculated, and forest plots were generated (see Figure 2). The heterogeneity of effects among the studies was small (I ² = 27%; Modified H ² = 0.37; tau ² = 0.0009) and was not statistically significant (Cochran’s Q (df) = 4.11(3), p = 0.25). However, random effects model was adopted to reflect the diversity of studies. The overall HR (95% CI) was 1.09 (1.03, 1.16). The null hypothesis of HR = 0 was rejected (p = 0.002). A one-unit increase in AL was associated with a 9% increased risk of cancer-specific mortality (HR = 1.09, Test of overall effect: z = 3.118, p = 0.002).

Figure 2.

Forest plot and meta-analysis of studies reporting hazard ratio estimates for overall cancer-specific mortality (n = 4).

Discussion

This systematic review is the first to provide a summary of evidence regarding the examination of allostatic load in cancer. Despite comprehensive article search and retrieval efforts, only 12 articles fulfilled the inclusion criteria. All were published within the last 7 years, which supports an emerging interest in assessing AL in cancer.

This review supports the AL score as an appropriate index of physiological burden, especially among individuals with cancer. This was evident specifically among survivors of breast cancer, prostate cancer, and childhood cancer, although with varying rigor. Comparison of biological toll of cancer between Black and White women was done in only one study which appropriately addressed the confounding factors. But its small sample size of black women with history of breast cancer (n = 35) may not be representative of the whole population of black women who survived breast cancer. Nevertheless, the study findings point to the need for well-powered studies that examine the impact of cancer on physiological regulation especially among minority populations. The significant association of pre-diagnostic AL with poorer breast cancer clinicopathologic features among Black women also point out the importance of further examination of factors that contribute to aggressive breast cancer phenotypes in this population. Since tumor grade and tumor size are important contributors of aggressive tumor biology which in turn is likely to be associated with increased mortality among Black women (DeSantis et al., 2019), inclusion of AL in early strategies for cancer control and prevention could help reduce the risk of cancer-specific mortality. These findings suggest that biological risk factors of cancer or effect of cancer on physiological processes are disparate. Social inequalities in cancer burden and differences in AL between Black and White populations in the USA, consistent with the “weathering” hypothesis, are already known (Geronimus et al., 2006; Jemal & Siegel, 2019; Newman & Kaljee, 2017; Thorpe et al., 2019). Allostatic load burden is also found to partially explain higher mortality among Blacks (Duru et al., 2012). Considering these reports and our review findings, an increased focus on physiological burden, long-lasting health consequences of cancer, and early identification of physiological dysregulation especially among the minority population is warranted.

An important aspect of the review findings was that although operationalization of AL differed across the studies, biomarkers representing cardiovascular, metabolic, and immune systems were largely included in computation of AL. This finding might be due to the fact that physiological measurements (biomarkers) of these 3 systems are easily available or/and accessible due to their frequent use and relevance in daily clinical practice. This is further supported by the observation that all the studies that recruited participants from clinical settings rather than using population databases used biomarkers from these 3 domains. The neuroendocrine biomarkers such as epinephrine, norepinephrine, cortisol, serotonin etc. are not routinely assessed in clinical practice apart for individuals presenting with specific disease states. Although the cluster comprising “non-clinical” measures of neuroendocrine and immune functioning is more predictive of mortality than the cluster comprising standard “clinical” measures of cardiovascular and metabolic function (Goldman et al., 2006), including the former cluster for calculating AL might be difficult in regular oncological clinical practice. Even though primary mediators are fundamental elements in the AL framework, a composite consisting of cardiovascular, metabolic, and immune system indicators with known clinically relevant thresholds could be more pragmatic in oncology practice. Considering the ongoing debate on uncertainty of using DHEA-S in AL estimation and questions on whether a neuroendocrine subcomponent of AL is reflective of chronic stress (Gersten, 2008; Loucks et al., 2008), further clarification and a consensus is required. Additionally, heart rate variability (HRV) as a measure of SNS activation, was included only in one study. Recent research suggests inclusion of HRV in AL estimation (Mauss et al., 2016; Viljoen & Claassen, 2017) and this area also merits further investigation.

Differences in risk-scoring is also worth further exploration. Some researchers might prefer using previously defined cut-off values of biomarkers to enhance empirical or clinical relevance. In such cases, use of standardized clinical cut-off values with attention to sex differences in biomarkers need to be used. On the other hand, those utilizing quartile criterion for assigning a risk score are interested in a data-driven partitioning of the sample, where membership in the upper or lower quartile represents exposure to more extreme system responses compared to the rest of the population and thus potentially at greater risk of disease pathology (Seeman et al., 2001). Since AL emphasizes the concept of deviation from homeostasis rather than diagnosis of individual disease (Hwang et al., 2014), the high-risk quartiles seem to be “statistically” correct method in a random cohort of cancer population but as the rigor of this method is driven by sample distribution, the criterion cut-off will depend on the sample size and individuals included in the study. Clinical significance of such data-driven cut-off for biomarkers need to be examined.

The variation in number and choice of biomarkers used in the studies make it difficult to conclude the ideal number and mix of biomarkers that are mandatory to estimate AL in cancer population. Using 10-12 biomarkers in clinical practice settings is cumbersome; this could explain the use of population-based databases to examine AL. The alternative of clinimetric criteria offer some hope in utility in clinical practice. Although these criteria have their unique limitations due to use of subjective measures, these clinimetric tools help put the use of biomarkers in a psychosocial context with particular reference to lifestyle modifications (Fava et al., 2019). It is worthwhile to note here that Seplaki et al (2005) demonstrated that choices regarding which biomarkers to include in a summary measure and how the measure is formed have only modest effects on the outcome; while including risk factors at both high and low tails and measures that preserve the continuous properties of the biological variables (e.g., the z-score) yielded stronger prediction of health outcomes.

A promising finding of the review is the significant associations between AL and cancer-specific mortality risk. Two studies in the review suggested that AL may not be as useful as a predictor for specific causes of death as for all-cause mortality and thus weaken the case for AL being a useful predictive tool for focused prevention measures. But it must be noted that one of these studies was done exclusively among Scottish population (Robertson et al., 2017) and the other examined AL as a mediator-moderator of association of DASH diet with cancer-specific mortality (Beydoun et al., 2019). This association warrants further research, considering that 3 studies with substantially large sample sizes reported positive association of AL with cancer mortality and the mini meta-analysis done by the review authors found a 9% increased risk of cancer-specific mortality with every unit increase of AL. Thus, it can be concluded that AL could be a powerful predictive instrument of cancer-specific mortality.

Limitations

Although methodological rigor was fairly good across the studies, all the studies that used a population-database have limitations unique to the secondary nature of the analyses and therefore warrant caution in interpretation. Biological markers utilized to estimate the composite AL score and participant characteristics are restricted to availability in the databases, thus risking selection or misclassification bias, despite multiple imputation methods. Additionally, there is an increased potential for residual confounding due to unmeasured or inadequately measured confounders, as the settings for assessments are not in the control of the researcher who is utilizing the secondary data. For instance, the data is not current and the existing socio-politico-cultural climate and health resources during data collection period could influence stress responses. Also, the biomarker data for NHANES respondents is only available for a single time point, preventing study of changes or trajectories of AL.

In all the cross-sectional studies that found association of AL with either an antecedent or an outcome, the temporal relationships between exposures cannot be determined. For instance, it cannot be determined if higher AL scores predisposed Black women to developing breast cancer or were a result of the disease (Parente et al., 2013). Similarly, we cannot draw conclusions about the potential reverse causality between posttraumatic overgrowth and AO (Ruini et al., 2015). It is important to assume that individuals taking anti-hypertensives, hypoglycemic agents, or other medications have already experienced physiological dysregulation; so the effects of medications need to be adjusted to better capture the underlying biomarker values (Geronimus et al., 2006). Only 4 of the 12 studies considered this aspect while estimating AL. Finally, all the biomarker samples were obtained from blood or saliva which are measures of acute stress. Biomarkers of chronic stress such as hair cortisol (Russell et al., 2012) were not used in any of the studies. One reason for this is that the studies using the population databases were limited by the data available in the respective databases (Mattei et al., 2010; Robertson et al., 2017) and none of them possibly had data on hair cortisol. Additional limitations related to collection and interpretation of hair cortisol levels could also be reasons why it was not used in the studies. Few aspects of hair cortisol analysis are still not fully understood, such as variability in hair cortisol levels related to genetic predisposition, exposure to chronic stress during early development, hair growth rate, length and types of hair, hair hygiene practices, and site of hair collection (Meyer & Novak, 2012; Russell et al., 2012; Wright et al., 2015). Another limitation in hair cortisol analysis is related to inter-assay variability and lack of agreed upon gold-standard measures or established reference values for age, sex, or ethnicity (Meyer & Novak, 2012; Russell et al., 2012; Wright et al., 2015). Additionally, the accumulation of hair cortisol levels represents a mean level of HPA activity across long time periods and does not provide a measure of diurnal variation (Meyer & Novak, 2012; Wright et al., 2015). Nevertheless, hair cortisol provides a new approach to assessing long-term HPA activity (Meyer & Novak, 2012) and has potential to fill the methodological void regarding the assessment of long-term cortisol (Gerber et al., 2012).

Although AL has shown great potential as an interdisciplinary tool for assessing cumulative health risk due to its theoretical complexity (Galen Buckwalter et al., 2016), certain limitations of the AL framework also need to be reflected upon. Debates surrounding the measurement of AL include which biomarkers to include, how to measure, combine, and weight them, and what statistical analytic techniques are appropriate. The questions of whether to represent AL biomarkers as continuous, standardized, categorical, or dichotomous variables and whether risk-defining cut-offs should be sample based, sex-specific, or at one or both extremes of the distribution or whether clinical criteria are more appropriate remain unresolved (Galen Buckwalter et al., 2016; Seeman et al., 2010). Additionally, biomarkers for allostatic load/overload are nonspecific mediators and do not provide information on the underlying individual experiential causes—a gap which is filled by the clinimetric criteria for AO (Fava et al., 2019). Also, specifying thresholds for biomarkers could seem counter-intuitive because of the expected use of AL for providing early warning signs for future adverse health outcomes (Seeman et al., 2010). Finally, while estimating AL, the choice of sample also needs to be carefully decided, as for most neuroendocrine biomarkers, a single one-time sample from blood or saliva might not be as useful as cumulative measure obtained from urine samples.

The heterogeneity in operationalization of AL and divergent antecedent and health outcome relationships hindered a meta-analysis of all 12 studies. Another limitation of this review is inclusion of studies available in English. The findings from studies published in other languages could have further strengthened the analysis and might offer more understanding regarding AL in cancer. Also, it is possible that additional studies on the subject have been published since the completion of this analysis. Nevertheless, this is the first review to report on AL in cancer, and its findings allow a precise evaluation of the current evidence in the AL literature.

Implications in Cancer

Since variability is common in the early stages of biomarker panel development (Gallo et al., 2014), the variability in operationalization of AL could be considered a reflection of the trajectory of advancement in use of AL construct in cancer. AL may prove to be a useful screening tool for high-risk individuals by highlighting early points to intervene, as long as it is estimated using empirically robust and accurate methods with attention to sex differences, circadian changes in biomarkers, and correct choice of sample (urine, saliva, hair, or blood). If the AL composite measure can help to identify a window period for intervening before tertiary outcomes occur, it holds a promise to potentially prevent premature cancer deaths. Estimating AL could be made more practical and clinically relevant by finding a definitive limited set of variables. For instance, hypothetically, a gold standard of subjective measures (clinimetric criteria) or a standardized cluster of specific biomarkers reflecting primary and secondary mediators (for instance, cardiovascular-metabolic-immune systems) with specified threshold values, will improve the clinical utility. Such an AL measure could then be used more frequently to evaluate efficacy of interventions in usual clinical settings. In a recent scoping review of interventions targeting AL as an outcome, the authors concluded that AL could be a biological outcome measure that is sensitive to change in response to interventions, and that future studies need to examine if AL serves as a mediator in the effects of the intervention (Rosemberg et al., 2020). AL being amenable to change opens up a promising area of research in cancer. AL could be included as an outcome in targeted intervention research which would additionally inform how body responds to cancer exposure. Following this, population-based preventive intervention program in communities could be established. Longitudinal studies could help to understand if AL predisposes individuals to cancer or whether it is a cumulative consequence of the cancer burden. Such studies would also help to examine AL trajectories to improve understanding cancer progression and susceptibility to death.

Conclusion

Allostatic load holds a significant promise in cancer research and practice, and ongoing examination regarding optimal standardized approaches to measure AL in cancer is warranted. A particular need while standardizing the approach is to consider the availability and usability of the biomarkers by practitioners in oncology. Once estimation method is standardized and causal direction between AL and cancer is well-understood, AL could be routinely incorporated as a screening tool or a health outcome in cancer.