A meta-analysis of the association between physical demands of unpaid domestic work and back pain among women

Abisola Modupe Osinuga (  abisola-osinuga@uiowa.edu ) University of Iowa BME: The University of Iowa Roy J Carver Department of Biomedical Engineering https://orcid.org/0000-0002-1418-1661 Chelsea Hicks University of Iowa: Department of Occupational and Environmental Health Segun E Ibitoye University of Ibadan Marin Schweizer university of Iowa: Department of Epidemiology Nathan Fethke University of Iowa: Department of Occupational and Environmental Health Kelly Baker University of Iowa: Department of Occupational and Environmental Health


Background
Back pain (BP) includes acute (less than 6 weeks), chronic (pain lasting more than 12 weeks), or neuropathic pain in the upper or lower back. BP is a leading contributor to disability and activity limitation and has been estimated to account for 37% of the global burden of musculoskeletal disorders (MSDs) [1].
In addition, low back pain (LBP) remains the leading cause of years lived with disability (YLDs) among adults, accounting for 64.9 million YLDs [2]. Although BP can be a self-limiting condition, the reoccurrence rate is high (approximately 60%) and individuals are likely to experience another episode within 3-6 months [2,3]. Limitation of activities because of LBP may result in loss of productive work time, increased medical expenditure, and further entrenchment of low-income individuals in poverty [4,5].
A systematic literature search was conducted to identify records published from January 1991 to March 2020 in PubMed, Embase, Web of Science, Scopus, and CINAHL, using search terms and keywords related to the population of interest, the exposure, and the outcome (" back pain" or "musculoskeletal symptoms"), without language restrictions. Google Scholar and the reference lists of relevant articles were searched for additional citations. This meta-analysis was performed and reported based on the Meta-analyses of Observational Studies in Epidemiology (MOOSE) criteria and in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist [37,38].

Eligibility Criteria.
Included studies were those that were published in peer-reviewed journals; assessed unpaid domestic work exposures; focused speci cally on women, or women represented at least 50% of study sample and gender-strati ed effects were presented; described at least one type of domestic task done by women; included women 18-59 years of age; and reported data needed to derive the measure of effect and the corresponding 95% CI. Studies that were anecdotal, case series, editorials and reviews were excluded. Studies were also excluded if they focused only on occupational (paid) domestic work; included only elderly (over age 65 years) or pregnant women; included a predominantly male sample; did not present gender-speci c effects; included acute musculoskeletal injuries (e.g., from slips/trips/falls or other traumatic events) or focused on MSDs in body areas other than the back (e.g., shoulder pain and carpal tunnel syndrome) as the outcome; included participants with BP secondary to a speci c disease (e.g., osteoporosis and cancer); described exposure based solely on the woman's role (e.g., "housewives") but not on PDDW; focused on caregiving for disabled persons; or did not report effect estimates separately for unpaid domestic work exposures.

Data Extraction.
Two reviewers (AO and CH) independently reviewed and abstracted information from potential studies after duplicate reports had been removed. Disagreements were resolved by consensus and discussion with a third reviewer (SI). The following information was abstracted: author, the year of publication, country of study, study design, proportion of the sample that included women, results strati ed by gender (yes/no/not applicable), at least one type of domestic tasks is speci ed (yes/no), mean age of women participants, study sample size, numbers in exposed and unexposed groups, de nition of BP, measures of domestic work exposures.

Assessment of Study Quality.
The Newcastle-Ottawa Scale for Observational Studies [39] was used to evaluate the risk of bias among the studies identi ed for inclusion. Some modi cations were made to the scale based on the study design (Supplemental Table 1), For example, criteria for exposure de nition were modi ed in the scale.
The assessment was conducted by two reviewers (AO and CH) independently. Disagreements were resolved by discussion or consultation with a third reviewer (SI). The scale, totaling 10 points, evaluates the risk of bias from three domains: the selection of the study sample (representativeness of sample/cases, adequacy of sample size, response rates, de nition and ascertainment of exposure and de nition of cases and controls), comparability of study groups (potential confounding variables), and conceptualization of outcome (outcome de nition and ascertainment, appropriateness of statistical measures). Quality scores were categorized as high (≥ 8 points), moderate (6-7 points), or low (≤ 5 points). For the risk of bias plots, the Robvis visualization tool was used to categorize and color-code each element in the three domains based on high risk, some concerns, and low risk of bias [40].

De nition of Outcome and Exposure Variables.
The outcome was de ned as either self-reported pain/discomfort in the back area (upper, lower, or nonspeci c) or care-seeking for BP. We included studies regardless of recall period used to ascertain selfreported back pain status. PDDW, which is the exposure variable, were expected to be assessed differently across studies because the dimension of work exposures is typically characterized in terms of frequency, intensity, and duration [41]. Therefore the exposure variable included those that categorized PDDW in terms of the magnitude of work (heavy or light PDDW), or assessed the duration of domestic work (time spent on domestic work per week/per day), or assessed the frequency/time spent working in awkward postures or history of forceful lifting.

Data Synthesis and Statistical Analysis.
The primary analytic objective was to estimate, using exposure and outcome data abstracted from each included study, a pooled estimate of association PDDW and BP. We also conducted subgroup analyses by country status (high versus low-and middle-income countries), by de nitions of exposure (time, frequency/intensity of work, and biomechanical factors), and focus on gender (women only versus studies with both male and female included). If a study examined all the types of PDDW exposure, only the biomechanical exposure was analyzed. Strati ed analyses were performed based on the study quality (i.e., high/moderate/low) and whether effect estimates were adjusted to control for confounding (yes/no).
Since most studies used a cross-sectional design, the OR was used as the measure of association. Measures of effect from included studies were pooled using the natural logarithm of the ORs (logOR), and the OR from each study was weighted by the inverse of its variance. Unadjusted ORs were used if adjusted ORs were not provided. When no effect estimate was given, the unadjusted OR was calculated directly from the abstracted information. A random-effects model was used to estimate the pooled OR and the 95% CI [42]. The data were pooled in Microsoft Excel, analyzed in the Review Manager (RevMan) version 5.3 program [43] and with the package 'meta' in R [44].
Heterogeneity among studies was examined by using the Cochran's Q test and quanti ed using the Higgins I 2 statistic [45]. We set the criterion for a statistically signi cant Cochrane Q test to p < 0.1. The degree of heterogeneity was de ned as low (I 2 < = 25%), moderate (26-50%), high (I 2 = 51-75%), and very high (I 2 > 75%). Publication bias was assessed using funnel plots and Egger's test [46,47]. The Leave-oneout method, and Baujat Plot, were used to investigate the effect of outliers and in uencers on the degree of heterogeneity. We conducted sensitivity analyses to examine the in uence of outliers and in uencers, studies with low quality scores, or those that did not adjust for confounders on the Pooled OR. Figure 1 shows the detailed results of the identi cation and study selection process. We retrieved 1,358 non-duplicate records through search of databases and references from relevant articles. Following review of titles and abstracts 1,311 studies were excluded based on the following: not relevant to the outcome of interest (disabilities, domestic violence, infectious diseases, surveillance data, or acute musculoskeletal injuries), not relevant to the exposures or outcome of interest, not relevant to the population of interest, or were reports, reviews, case series or editorials. Forty-seven full text articles were subsequently assessed for eligibility and 36 were excluded as shown in Fig. 1. And Supplemental Table 4.
Seven studies recruited only women [25,48,50,[52][53][54]57], three studies either had a sample that included at least 50% women or reported gender-strati ed results [49,51,55], and one large-scale study did not specify the number of women recruited but strati ed results by gender [56]. All studies speci ed common domestic tasks done by women (e.g. cleaning, cooking, child caregiving). Four studies were on full-time housewives/homemakers [25,50,54,57] while the remaining eight included women with paid employment.
Three studies categorized PDDW as number of hours per week performing domestic tasks [50][51][52] four studies categorized exposure as biomechanical (lifting, carrying, working in static and awkward postures) [25,54,55,57] and the rest categorized PDDW based on intensity (heavy or light) and frequency of work [48,49,53,56]. In all included studies, exposure information was ascertained by self-report (i.e. questionnaire). Eight out of eleven studies included in this review were on low back pain while the remaining three were on chronic back pain, care-seeking for low back pain and upper back pain. The only case-control study de ned outcome as care-seeking for low back pain [52] while the rest used selfreported (7 days, 1, 3, or 12 months) BP. All studies except one reported ORs adjusted for relevant confounders such as age, education, occupation, income and psychosocial factors of paid work.

Quality Assessment.
Using the Newcastle Ottawa Scale, six studies were categorized as high quality (8 points and above), four as moderate quality (6 and 7 points), and one as low-quality (5 points). The Robvis visualization plot in Fig. 2 shows that most studies have low risk of bias in most of the sub-domains except for assessment of non-respondents (a subdomain of selection of study participants). The quality assessment table and Robvis plot are depicted in Supplemental Table 2 and Fig. 2, respectively.

Overall association between domestic work demands and BP.
The pooled odds ratio from eleven studies showed that high PDDW, characterized as long duration, high frequency and intensity, or high biomechanical demand (awkward posture and lifting heavy objects), was signi cantly associated with BP among women (OR = 1.63; 95% CI = 1.30-2.04; Fig. 3). However, there was substantial heterogeneity among the included studies (Cochrane Q-test p-value < 0.01; I 2 = 70%; Fig. 3).

Outlier and In uential Analysis.
The Baujat diagnostic plot (Supplemental Fig. 1) showed that Alzaharani et al. [49] contributed to the overall heterogeneity to the greatest extent and had the most in uence on the overall pooled OR, likely due to the large sample size compared to the other studies. In uential analysis, using the leave-one out method (Supplemental Fig. 2), also revealed that Alzaharani et al. [48] was the predominant source of heterogeneity. The lowest I 2 value (16%) was observed when the Alzaharani et al. [49] was removed from the analysis. Omitting Habib and Rahman [53]the study with the greatest effect size (OR = 6.00; 95% CI = 1.57-22.88), did not meaningfully in uence the pooled effect size (OR = 1.57; 95% CI = 1.26-1.95; Supplemental Fig. 2) or the degree of heterogeneity (I 2 = 69%).

Strati ed analysis.
Compared to the overall pooled OR of 1.63, slightly lower effect sizes were observed when considering only studies categorized as high quality (OR = 1.54 95% CI 1.18-2.02; I 2 = 80%; N = 6 studies, Supplemental Fig. 3) and when omitting the one study [53]

Publication Bias.
The funnel plot, a plot of the odds ratios against the standard errors from each included study appears symmetrical, suggesting limited evidence of publication bias (Supplemental Fig. 4). This conclusion is also supported by the result of the Egger's test (p = 0.15).

General Discussion
To the best of our knowledge, this is the only synthesis of observational studies that assessed the association between PDDW and BP while comparing estimates based on country status and standardized dimensions of physical work exposures (frequency, duration and magnitude). Our results demonstrated that women who perform high PDDW, characterized as long time spent on work, frequency and magnitude (heavy workload) and the presence of biomechanical risk factors (activity conducted in non-neutral posture) have a higher odds of BP compared to women who perform low PDDW. The magnitude and direction of the pooled estimates from the overall and subgroup analyses did not change meaningfully when the major source of heterogeneity was omitted and when analyses were restricted to high quality studies.
Standard operationalization of biomechanical exposure comprises three main dimensions; level or magnitude, repetitiveness or frequency, and duration of work. Exposures in the included studies were described in self-reported magnitudes, frequency and duration of domestic tasks such as carrying, lifting, pushing as well as self-reported frequency or duration of working in awkward postures and lifting. Biomechanical loading from manual material handling activities and working in awkward postures are established risk factors for both recurrent and chronic back pain [34]. This agrees with our results, speci cally that the pooled OR from studies in which exposures were de ned based on biomechanical characteristics of domestic work was greater than the pooled ORs from studies in which exposures were de ned in more generic terms (e.g., time spent performing domestic work).
We did not identify any prospective studies to include in the analyses, and a previous review of both occupational and non-occupational risk factors for LBP did not include domestic work exposures [34]. We are unaware of other studies assessing the association between PDDW among women speci cally, although one review of prospective studies found that women in the general population had a higher prevalence of low back pain compared to men [11]. The gender differences in prevalence of BP could be due in part to women in many regions of the world being disproportionately exposed to physically demanding domestic work. The results of this study suggests that the physical demands of unpaid domestic work and its musculoskeletal health risks should be more recognized and examined in occupational research, especially in LMICs where millions of low-income women are exposed to strenous daily work conditions that are comparable to work in occupational environments.
A prior systematic review of 35 studies assessed the association between occupational lifting and low back pain and concluded that it is unlikely that occupational lifting is independently associated with LBP [13]. Another systematic review found no signi cant relationship between free-living physical activity (classi ed as leisure-time physical activity) and non-speci c LBP [58]. The result of the former review may be different from ours because we included studies that examined non-neutral postures as biomechanical risk factors rather than just the activity of lifting. Also, our study assessed domestic physical activity among women in the household, not workplace physical activity. For the latter prior review, domestic physical activity was classi ed as a component of leisure time physical activity, which could have in uenced the exposure-outcome relationship.
Our results revealed a higher prevalence of BP in LMICs when compared to higher income countries. This agrees with other reviews studies where the prevalence of LBP among studies from LMICs was higher than those from high income countries although the studies did not consider gender-speci c effects [59,60]. The higher odds of BP from pooled effect estimates from LMICs indicate that the impact of domestic work on women's health may be more severe in LMICs. The biomechanical demands of domestic work on women may be elevated where water infrastructure is lacking, or unreliable and water carrying/porterage is common [61]. Water carriage has long-and short-term impacts on musculoskeletal health of women [29].
In addition to water insecurity and carriage, daily domestic tasks such as caregiving activities and manual food processing could increase the risk of BP [27]. Women from developed economies may have less strenuous domestic work demands than low-income women in LMICs as a result of differences in social/cultural expectations, better bargaining power due to higher paid income, increased use of mechanized household devices (e.g. dishwashers, laundry machines), and more social support/spousal involvement in performing domestic tasks. Similarly, middle-to-high income women in LMICs who have access to similar social and environmental resources might also face lower domestic work health risks.
These differences in domestic work demands partially explain why domestic work has traditionally been regarded as a 'non-occupational or leisure-time' physical activity with protective effects in published studies and reviews from developed countries [33][34][35].

Methodological considerations: Study Inclusion and Classi cation
The current study used well-de ned inclusion and exclusion criteria to ensure that studies containing the intended target population (women) were selected. We excluded studies focused on women taking care of persons with disabilities and elderly women out of concern their level of exposure would not be generalizable to the overall population and because there is causal association between advanced age and low back pain [11]. We also excluded studies that did not fully provide exposure and outcome information to reduce the threat of differential misclassi cation of exposure or outcomes. We included studies examining acute and chronic back pain. We believe our inclusion/exclusion criteria were appropriate given that prior meta-analyses of non-occupational exposures and back pain have not assessed the impact of the domestic work or focused on gendered effects.

Strengths and Limitations
The strengths of this review include: (1) extensive literature searches and inclusion of a study not published in English; (2) included studies were mostly of moderate to high quality and adjusted for relevant confounders; (3) identi cation of major sources of heterogeneity; (4) absence of publication bias; (5) Estimation of gender-speci c association of domestic work with back pain. Almost all (six out eight) studies that sampled women engaged in paid work adjusted for physical demands of paid work and other relevant demographic factors during data analysis.
Several limitations should be considered when interpreting the ndings from this study. Since the exposure and outcome information in most of the included studies were obtained using self-report methods, the results could be subject to reporting bias. Further, the temporal relationship between exposure to PDDW and BP cannot be established because most studies were cross-sectional, and we did not nd any prospective cohort study to include in this meta-analysis [34]. Other potential sources of bias were those relating to study selection such as adequacy of sample size, response rate reporting, and comparability between respondents and non-respondents.
Furthermore, few of the included studies reported information regarding the validity of the instrument used ascertain domestic work exposures. In addition, there were considerable methodological differences across the included studies in the measurement of domestic work exposures, which may have led to misclassi cation of domestic work demand in our analyses [62]. Likewise, the de nition, presenting symptoms, severity and period prevalence (ranges from one week to a year) of back pain differed across studies. These variations in exposure and outcome measurements across studies may have impacted both the magnitude and precision of our pooled OR estimates. Finally, since most studies ascertained both exposure and outcome by self-report, common method bias may have created the appearance of an association [63].

Implications and recommendations for research
Results of the current study suggest an association between the phyiscal demands of unpaid domestic work and BP, particularly among women in LMICs. Back pain is not typically a priority for mitigation because it is neither life-threatening nor as dangerous as other diseases prevalent in LMIC, yet it can cause long-lasting disability and declines in wellbeing and economic opportunity [59,60]. Presently, gender-speci c research investigating domestic exposures and musculoskeletal pain among women are limited [16,22]. Presenting data on how domestic work affects the musculoskeletal health of women will be important in designing future interventions (behavioral, infrastructural and ergonomic) that can reduce domestic work burden.
Most published large-scale studies that have assessed domestic work impact on musculoskeletal disorders mostly ascertain exposures using a woman's work role (housewives) and self-report of exposure [33,64,65]. Relying solely on self-report information of exposures collected at one point in time from participants may be imprecise or lead to misclassi cation of exposure especially in routine daily activities such as domestic work [66]. Likewise, more information should be collected on pain experiences beside 'absence or presence of pain' to correctly ascertain the presence and severity pain in future studies [67,68]. Future research should use objective measures of exposure, such as observation or instrumentbased tools, to quantify the physical demand of domestic work so that exposure information is rigorous and can be standardized across studies [69]. Using a pain ratings scale, visual diagrams, speci c recall period and probes to evaluate type of pain symptoms or level of impairment will be invaluable in reducing heterogeneity across studies, improving stability of estimates and generalization of ndings. More longitudinal studies are needed to estimate the day-to-day variances in domestic work, and to quantify the physical demands on women's bodies from domestic work, before we can fully understand its causal relationship with BP.
Policymakers and labor organizations should put more concerted efforts in recognizing unpaid domestic work as an occupation with health risks like those experienced in paid domestic occupations, which will aid in meeting global sustainable development goals for gender equity and women's health.

Conclusion
Results of the current study suggests that PDDW is associated with BP, especially among women from LMICs. Large prospective studies in LMICs are needed to critically examine or establish the relationship between domestic physical activity and BP.

Availability of data and Materials
The data analysed were abstracted from published articles, other data generated are available in the supplemental le.

Competing Interests
The authors declare that they have no competing interests.