Association between job control and time to pregnancy in a preconception cohort

Abstract

Objective:

To evaluate associations between low job control (operationalized as job independence and freedom to make decisions) and time-to-pregnancy. Low job control, a form of workplace stress, is associated with adverse health outcomes ranging from cardiovascular disease to premature mortality; few studies have specifically examined its association with reproductive outcomes.

Design:

We used data from Pregnancy Study Online (PRESTO), an internet-based preconception cohort study of couples trying to conceive in the U.S. and Canada. We estimated fecundability ratios (FRs) and 95% confidence intervals via proportional probabilities regression models, adjusting for sociodemographic and behavioral characteristics.

Subjects:

Participants self-identified as female, were aged 21–45 years, and reported ≤6 cycles of pregnancy attempt time at enrollment (2018–2022).

Exposure:

We assessed job control by matching participants’ baseline self-reported occupation and industry with standardized occupation codes from the NIOSH Industry and Occupation Computerized Coding System, then linked codes to O*NET job exposure scores for job independence and freedom to make decisions.

Main outcome measure:

Our main outcome measure was fecundability. Participants completed self-administered questionnaires at baseline and every 8 weeks for up to 12 months or until reported pregnancy, whichever occurred first.

Results:

Among 3,110 participants, lower job independence was associated with reduced fecundability. Compared with the fourth (highest) quartile, corresponding to the most job independence, FRs (95% CI) for first (lowest), second, and third quartiles were 0.92 (0.82–1.04), 0.84 (0.74–0.95), and 0.99 (0.88, 1.11), respectively. Lower freedom to make decisions was associated with slightly reduced fecundability (first versus fourth quartile: FR=0.92; 95% CI: 0.80–1.05).

Conclusion:

Lower job control, a work-related stressor, may adversely influence time-to-pregnancy. As job control is a condition of work (i.e., not modifiable by individuals), findings may strengthen arguments for improving working conditions as a means of improving worker health, including fertility.

Capsule:

Workplace stress, specifically low control over work time, may be a risk factor for subfecundity among working women who are trying to conceive.

Introduction

Low control over work is a well-established dimension of occupational stress (1). Low job control is characterized by both one’s limited ability to make independent decisions about how and when work is done (low decision authority) and limited ability to exercise creativity in one’s work (low skill discretion). It has been linked with adverse health outcomes, including premature mortality (2), cardiovascular disease (3), and hypertension (4). Job control is hypothesized to impact these outcomes through multiple mechanisms. Biological pathways include dysregulation in immune and inflammatory markers (5), and behavioral pathways include coping-related responses (e.g. alcohol consumption, sleep) (6).

Studies have examined the influence of occupational exposures on couples’ fertility, as measured by fecundability (probability of conception with unprotected intercourse during a given menstrual cycle) and time to pregnancy (TTP) (number of months or menstrual cycles until achieving pregnancy). In general, preconception occupational exposure to lead, pesticides, and antineoplastic agents are associated with longer TTP (7, 8). Results are less consistent for non-chemical occupational exposures, e.g. lifting heavy loads (9), night shifts, and long work hours (10, 11). Although the latter factors have been associated with menstrual cycle irregularity, non-chemical occupational exposures are not consistently associated with TTP or infertility. Next-generation occupational health frameworks, such as Total Worker Health^®, posit that occupational health risk factors go beyond chemical and biomechanical hazards to include psychosocial and organizational exposures (12). Although such exposures are frequently assessed as risk factors for chronic disease, injury, and mental health outcomes, they are rarely considered in association with reproductive health outcomes.

General life stress, whether measured by self-report or biomarkers, has been associated with longer TTP in prospective cohort studies (13, 14). Although work is a major source of life stress among working adults (15), work-specific stress, and psychosocial working conditions more broadly, have seldom been evaluated as risk factors for longer TTP. Thus, we evaluated to what extent job control was associated with fecundability among working women in a preconception cohort. We hypothesized that job control would be inversely associated with TTP, such that those with less job control would take longer to conceive than those with high control.

Methods

Study population:

Pregnancy Study Online (PRESTO) is an ongoing web-based preconception cohort of pregnancy planners in the United States and Canada (16). Participants first completed a screening questionnaire with questions about age, sex, pregnancy status, use of contraception or fertility treatments, partner status, and pregnancy attempt time to date. Eligible participants were 21–45 years old, lived in the United States or Canada, self-identified as female, reported having a male partner, and were attempting to conceive without the use of fertility treatment. If eligible, they were immediately sent a baseline questionnaire with questions about demographic, social, behavioral, and reproductive factors.

Between June 2013 and May 2022, 16,434 eligible participants completed baseline questionnaires. We excluded participants from TTP analyses for: >60 days between screening questionnaire and baseline questionnaire (n=46), last menstrual period date >6 months before baseline questionnaire (n=164), no prospective menstrual period dates or pregnancy over follow-up despite completing follow-up questionnaires (n=41), and >6 cycles of pregnancy attempt time at study entry (n=3,195). Because we added detailed questions on occupation and industry to PRESTO baseline questionnaires in June 2018, we excluded those completing their baseline questionnaire before June 2018 (n=5,872). These criteria yielded 7,114 participants for analysis. See “Exclusions for occupational analyses” below for further exclusions.

Exposure.

We assessed job control via a well-established modeling approach (17, 18). Occupational information was reported as a job title (free-text field) and industry (from a list of 52 standard Census industry categories) (19). We employed the NIOSH Industry and Occupation Computerized Coding System (NIOCCS) to translate industry and occupation fields into standardized occupation codes (SOCs) and industry codes (20). Next, we used the NIOCCS SOCs to link PRESTO participants’ occupations to modeled occupational exposures from O*NET. O*NET is a job exposure matrix maintained by the U.S. Department of Labor (21). Each of 873 SOCs is scored 1–100 on a range of physical, social, task-oriented, and organizational exposures typically associated with that occupation.

The two O*NET exposures used for this analysis were job independence (ability to determine how and when to complete work) (22) and freedom to make decisions (ability to make work decisions without others’ input or permission) (23). In general, higher values of job control are considered less stressful (24) and are more prevalent in higher-status jobs (25). An example of a high-independence job is a university professor; a low-independence job is a restaurant server. Each dimension is scored on a scale of 1–100, For analysis, we created quartiles based on distribution of scores within the sample for each dimension of job control.

To assess robustness of NIOCCS coding, we compared NIOCCS-generated industry codes with self-reported industry code; 83% were perfect matches, and 3.5% of participants had unexpected occupation-industry combinations. To assess accuracy of O*NET exposures, we compared a question on PRESTO’s baseline questionnaire (“Are you exposed regularly (daily or almost every day) to environments with temperatures above 77°F during work or home activities?”) with O*NET rating of “exposure to extreme cold or heat at work.” Of 7% of PRESTO participants who self-reported frequent exposure to high temperatures at work or home, mean O*NET occupational heat exposure score was 21.06 (SD 22.49) (scale: 0–100); among 93% who did not report high temperature exposure at work or home, mean O*NET heat exposure score was 13.51 (SD 14.18).

Outcome.

Our outcome was time to pregnancy (TTP), calculated with data from baseline and follow-up questionnaires. At baseline, participants reported last menstrual period date (LMP), usual menstrual cycle length, and number of menstrual periods they had been trying to conceive pre-enrollment. At each follow-up, participants again reported LMP and whether they had conceived since their last questionnaire. For those with irregular cycles, usual cycle length was calculated using baseline LMP and subsequent LMPs on follow-up questionnaires. Attempt time was calculated based on total discrete cycles at risk of pregnancy.

Covariates.

We adjusted for participants’ baseline characteristics: age (years: <25, 25–29, 30–34, 35–39, ≥40 ), education (high school; some college; college graduate; graduate school), income level (U.S. dollars: <$50,000; $50,000–99,999; $100,000–149,999; ≥$150,000), intercourse frequency (<once/week; once/ week; 2–3 times/week; ≥4 times/week), self-identified race and ethnicity (non-Hispanic Black; non-Hispanic White; Hispanic or Latina; non-Hispanic Asian/Pacific Islander; non-Hispanic mixed race or other), geographic region (Northeastern US; Southern US; Midwestern US; Western US; Canada), year of study entry (2018, 2019, 2020, 2021, 2022), and taking active steps to improve chances of becoming pregnant, e.g. timing intercourse or tracking ovulation (yes, no).

Exclusions for occupation analyses.

Of 7,114 eligible participants, we excluded those not working for pay at study baseline (n=924). Of 6,220 remaining, 5,990 (95%) reported both occupation and industry, necessary for NIOCCS coding. Of these, 4,478 (76%) were rated by NIOCCS as having >90% confidence in match strength for occupational code. Following standard practice (26), we retained only these high-confidence matches. Of 4,487 participants with high-confidence NIOCCS-generated SOCs, 3,110 (69%) had SOCs with corresponding O*NET data and were included in analyses. See Supplemental Table A for comparisons of those with and without occupational exposure data.

Data analysis.

Following baseline surveys, participants responded to bimonthly follow-up surveys for up to 12 menstrual cycles or until reported pregnancy, to update exposure data and pregnancy status. Participants contributed cycles at risk to analysis until reported pregnancy or a censoring event, whichever came first. Prior to multiple imputation (see below), these censoring events included 12 cycles of attempt time (n=257), no longer trying (n=68), initiation of fertility treatment (n=215), loss to follow-up (n=194), and still participating at the time of dataset creation (n=194).

We used proportional probabilities regression models to estimate fecundability ratios (FRs) and 95% confidence intervals (CIs) for associations between quartiles of exposure (job independence, freedom to make decisions) and fecundability. We constructed separate models for each exposure. FR is average per-cycle probability of conception in exposed participants versus those in the reference group (here, those in the fourth quartile of exposure). Inclusion of indicator variables for “cycle at risk” in models account for expected fertility decline in the analytic population during follow-up. A FR<1 is interpreted as longer time to pregnancy versus the reference group (lower fecundability); FR >1 is interpreted as shorter time to pregnancy versus the reference group (higher fecundability). We first modeled unadjusted associations, then adjusted for age, education, and income to evaluate whether any observed patterns might be attributable to socioeconomic differences in exposure distribution. Final models adjusted for age, education, income, intercourse frequency, race and ethnicity, geographic region, year of study entry, and actively trying to improve chances of pregnancy.

We used fully conditional specification (FCS) methods to multiply impute values for missing data on exposures, covariates, and pregnancy status (27). We created twenty imputed data sets with SAS PROC MI and combined coefficient and standard error estimates from datasets using SAS PROC MIANALYZE (28). Missingness ranged from <0.1% (race/ethnicity, education) to 3.0% (income); there were no missing values for age. We assigned participants without follow-up information (n=590, 19%) one cycle of observation and multiply-imputed their pregnancy status (yes vs. no).

Sensitivity analyses.

We conducted several sensitivity analyses. First, we used splines to model job independence and freedom to make decisions continuously. We restricted analyses to those working ≥30 hours per week (to capture sufficient “dose” of exposure; 83% of sample retained) (Supplemental Table B) and to those who had had ≤3 cycles of attempt time at baseline (88% of sample retained; Supplemental Table C). We also stratified by parity at study entry (nulliparous vs. parous) (Supplemental Table D).

Ethics approval:

PRESTO was approved by the Boston University Medical Campus Institutional Review Board.

Results

The analytic sample included 3,110 participants, who contributed 1,813 pregnancies during 10,948 menstrual cycles of attempt time (Table 1).

Table 1:

Baseline characteristics of 3,110 PRESTO participants (2018–2022). All characteristics, with the exception of age, were standardized to age distribution of cohort at baseline.

	Overall	Quartile of job independence
		First (lowest) (n=806)	Second (n=758)	Third (n=768)	Fourth (highest) (n=778)
Age, years (mean/SD)	30.2 (3.9)	30.1 (3.8)	30.3 (4.1)	30.8 (3.9)	29.8 (3.8)
Education,%
High school or less	3.6	4.3	6.7	2.9	0.1
Some college	18.5	19.3	24.4	17.0	13.9
College graduate	35.9	38.5	33.0	23.3	49.9
Graduate school	42.1	38.0	36.5	56.9	35.6
Annual household income US dollars, %
Less than $50,000	11.6	12.3	178	11.8	5.2
$50,000–99,999	34.0	30.7	37.2	31.7	37.0
$100,000–149,999	29.5	28.8	24.6	28.8	35.5
≥$150,000	24.9	28.2	20.4	27.8	22.3
Intercourse frequency, %
Less than once per week	22.0	233	23.5	21.3	20.1
1 time per week	19.7	20.6	18.2	20.8	18.6
2–3 times per week	44.9	41.9	44.3	45.2	48.0
≥4 times per week	13.4	14.3	14.0	12.7	13.3
Trying methods to improve chances of pregnancy
Yes	82.6	81.4	80.8	84.6	83.2
No	17.4	18.6	19.2	15.4	16.8
Race and ethnicity, %
Hispanic or Latina	6.7	6.9	7.4	6.6	5.6
Non-Hispanic Black	2.0	1.1	2.1	2.8	2.2
Non-Hispanic White	84.6	84.2	83.2	84.7	86.4
Non-Hispanic Asian/Pacific Islander	2.0	3.0	1.5	1.3	1.9
Non-Hispanic Other or mixed race	4.7	4.8	5.8	4.7	3.9
Region
Northeast United States	19.7	19.3	17.6	22.0	19.6
Southern United States	22.8	24.0	23.7	22.8	20.1
Midwest United States	23.7	19.9	23.9	25.0	26.5
Western United States	16.6	19.4	16.7	15.6	17.9
Canadian	17.3	18.4	16.7	15.6	17.9

When we examined distribution of key demographic characteristics by quartiles of job independence, we observed a socioeconomic gradient. This pattern is expected; job control typically increases with occupational prestige. Lower education (some college or less) was more prevalent among those in the lowest quartile (26% of participants) and second quartile (31% of participants) of job independence versus the highest quartile (14% of participants). While we did observe patterning by income, it was less consistent, with 17% of those in the lowest quartile making under $50,000 per year versus 12% in the second quartile, 18% in the third quartile, and 5% in the highest quartile. We saw little evidence of racial or ethnic disparities in exposure distribution.

When comparing characteristics of participants with (n=3,110) and without (n=3,110) full exposure assessment, we found few differences (Supplemental Table A). Those without exposure data were slightly less likely than those with exposure data to have a college degree (32% versus 36%), but slightly more likely to have some graduate school or more (49% versus 42%). Using life table methods, the percentage of participants reporting a pregnancy during 12 cycles of follow-up did not vary by exposure assessment (59% versus 58%).

Next, we examined distribution of O*NET scores (range: 1–100, 1=lowest exposure, 100=highest exposure) for both operationalizations of job control. For job independence, the median was 76 (IQR: 70 to 92; range 48 to 95); for freedom to make decisions, the median was 82 (IQR: 76 to 88; range 38 to 99). Although the theoretical range for each exposure is 1–100, within O*NET, the true range for job independence among all jobs indexed in O*NET is 44 to 96, and for freedom to make decisions is 33 to 100.

For job independence, we observed a J-shaped relationship between job independence and fecundability in unadjusted models (Table 2). While the magnitude of associations was attenuated after adjusting for age and socioeconomic factors and then for additional social and demographic characteristics, the J-shaped association remained. In fully-adjusted models, compared with those in the highest independence quartile, those in the first (lowest) quartile had FR=0.92 (95% CI 0.82, 1.04), in the second quartile a FR=0.84 (95% CI 0.74, 0.95), and in the third quartile a FR=0.99 (95% CI 0.88, 1.11). This pattern was also observed in spline models (Figure 1). We observed few differences in models adjusting for age and socioeconomic factors (education, income) only, compared with models adjusting for all factors.

Table 2:

Association between job control scores and fecundability.

				Unadjusted	Adjusted for age, income, education	Fully-adjusteda
Exposure	N participants	N pregnancies	N cycles	FR (95% CI)	FR (95% CI)	FR (95% CI)
Job independence quartile
1 (lowest)	806	471	2,869	0.90 (0.80, 1.01)	0.91 (0.81, 1.02)	0.92 (0.82, 1.04)
2	758	380	2,729	0.79 (0.69, 0.89)	0.82 (0.73, 0.94)	0.84 (0.74, 0.95)
3	768	487	2,737	0.99 (0.88, 1.11)	0.98 (0.87, 1.10)	0.99 (0.88, 1.11)
4 (highest)	778	475	2,613	1.0 (ref)	1.0 (ref)	1.0 (ref)
Freedom to make decisions quartile
1 (lowest)	755	369	2,573	0.81 (0.72, 0.93)	0.91 (0.80, 1.05)	0.92 (0.80, 1.05)
2	620	361	2,245	0.89 (0.79, 1.02)	0.93 (0.82, 1.06)	0.94 (0.83, 1.07)
3	984	593	3,435	0.98 (0.88, 1.09)	1.01 (0.90, 1.13)	1.02 (0.90, 1.14)
4 (highest)	751	490	2,695	1.0 (ref)	1.0 (ref)	1.0 (ref)

^aFully adjusted models account for education, age, income, race and ethnicity, intercourse frequency, year of study entry, geographic region, and trying to improve chances of pregnancy

Figure 1:

Spline models of continuous associations between (left panel) job independence and fecundability, and (right panel) freedom to make decisions and fecundability. Spline models are adjusted for education, age, income, race and ethnicity, intercourse frequency, year of study entry, geographic region, and trying to improve chances of pregnancy. Four knots are placed at the 25^th, 50^th, 75^th, and 90^th percentiles of the exposures. Reference group is the maximum value of (left panel) job independence and (right panel) freedom to make decisions.

We then examined associations between quartiles of freedom to make decisions and fecundability (Table 2). In unadjusted models, we observed a monotonic association between freedom to make decisions and fecundability, with lower fecundability at lower levels of freedom to make decisions. We observed moderate attenuation of the observed effects, especially in the lower quartiles, after adjusting for age, education, and income, and little additional attenuation after additional adjustment for other covariates. In fully-adjusted models, compared with those in the highest-freedom quartile, those in the lowest quartile had a FR=0.92 (95% CI 0.80, 1.05), the second quartile a FR=0.94 (95% CI 0.83, 1.07), and the third quartile a FR=1.02 (95% CI 0.90, 1.14). Spline models indicated a similar pattern (Figure 1).

When restricting to participants with ≤3 cycles of pregnancy attempt time at study entry (Supplemental Table B) or to participants working at least 30 hours/week (Supplemental Table C), associations largely mirrored our main findings, though confidence intervals were wider than in our main results. We also stratified by parity (Supplemental Table D). Here, we noted that among parous women, the J-shaped relationship with fecundability was more pronounced (FR for second quartile=0.74, 95% CI 0.59, 0.94) than among nulliparous women (FR=0.88, 95% CI 0.76, 1.03).

Discussion

We evaluated associations between job control (job independence and freedom to make decisions) and fecundability in a preconception cohort. Both operationalizations of low job control had modest associations with longer TTP, with somewhat stronger findings for low job independence than for low freedom to make decisions. Results were partially, but not fully, attenuated after accounting for to socioeconomic differences in exposure distribution. Although other studies have evaluated relationships between physical, chemical, and organizational exposures at work and time to pregnancy, this is one of the first analyses to evaluate associations between the psychosocial work environment and fecundability.

Potential explanations.

Overall measures of perceived stress (non-workplace-specific), especially during the follicular phase of the menstrual cycle, have been associated with reduced fecundability among women (13, 29). Potential explanations in those studies include stress-associated changes in hormones related to ovulation (30). However, few studies of fecundability have disaggregated life stress into specific domains, such as work.

Low job control has long been considered a core component of occupational stress, along with high psychological demands and low support (1, 24). Low job control is characterized by inability to make decisions about how and when to do one’s work. Here, we used two operational definitions of job control. The first, job independence, is ability to develop one’s own ways of accomplishing tasks, guiding oneself without supervision, and self-reliance at work (22). The second, freedom to make decisions, is ability to work without supervision (23).

Job control could be associated with fecundability for several reasons. Those who can make job-related decisions without consulting a supervisor might be more able to schedule doctor’s appointments or optimally time intercourse. We did find slightly lower intercourse frequency among those in the lowest-independence quartile (56% two or more times per week) compared with the highest-independence quartile (61.3% two or more times per week). However, adjusting for intercourse frequency did not attenuate job control-fecundability associations once accounting for education and income.

Biologically, the same hormonal mechanisms underpinning overall stress-fecundability associations could partially explain associations between job-specific stress and fecundability. Stress-related pathways may be supported by our finding that associations between job control and fecundability were stronger among parous than nulliparous participants. Work-family strain is associated with health outcomes including sleep and cardiometabolic risk factors (31). Work-family strain was amplified for working parents (particularly with young children) during the first years of the COVID-19 pandemic (32), which occurred during our study period. This timing could have intensified overall life stress among parous women, potentially contributing to subfecundity.

Alternative explanations for findings.

There may be non-causal explanations for findings. We did observe gradients in the distribution of job independence by education and, to a lesser extent, by income: among those in the lowest quartile of independence, 25% had some college or less; in the top quartile, 14% had some college or less. In prior studies, low socioeconomic status and lower income have been associated with reduced fecundability (33, 34). Socioeconomic gradients in a range of health outcomes (e.g., coronary heart disease, musculoskeletal disorders) have been partially attributed to low job control (35, 36). Although we adjusted for education and income, there may still be residual confounding by other measures of socioeconomic status or other work-related exposures.

Strengths and limitations.

Our study has limitations, namely risk of exposure misclassification. We assigned exposure values for job control based on standardized occupation codes—codes themselves generated from free-text fields using natural language processing—potentially introducing exposure misclassification at two points and compounding random error. We would expect misclassification to be non-differential with respect to the outcome (time to pregnancy), biasing results towards the null.

PRESTO is composed of couples planning a pregnancy and choosing to enroll in a preconception cohort. While prior preconception studies have found that Internet-based recruitment does not induce selection bias (37), couples planning pregnancy and enrolling in a preconception cohort may be different than those who are not. Specifically, PRESTO participants have higher socioeconomic status than the overall population, which is associated both with higher job control (35) and greater fecundability (33) than those of lower socioeconomic status. Accordingly, the range of exposure values was somewhat constrained in this cohort; for job independence, the mean was 76 and interquartile range 70–92 on a scale theoretically ranging from 1–100 and in actuality ranging from 33 to 99.

Modeled occupational exposures may have limitations for psychosocial occupational exposures; in order for an event to be stressful, the exposure (potential stressor) must both occur, and must also exceed individuals’ coping resources (38). Modeled exposures, like ours, only measure typical exposure and do not account for individual stress tolerance, desire to make one’s own decisions about work, or variation in stress perceptions within a given job (39). We would expect such elimination of within-job heterogeneity to introduce exposure misclassification that is non-differential with respect to the outcome. Despite these limitations, modeled psychosocial exposures at work are associated with outcomes including mortality (2), mental health (40), and musculoskeletal disorders (41).

Strengths stem from PRESTO’s prospective design. Occupation and industry were self-reported at study baseline, allowing us exposure assessment in close proximity to the beginning of attempting to conceive. We limited our analytic sample to those with fewer than six cycles of attempt time at study entry, and in sensitivity analyses, those with three or fewer cycles, reducing likelihood of selection bias and bias due to reverse causation (longer time to conception changing people’s work-related behaviors and exposures).

Conclusions.

Low job control may be a risk factor for longer time to pregnancy. As job control is a condition of work (i.e., not modifiable by individuals, but potentially modifiable through work redesign and other organization-level interventions) (42), findings may strengthen the case for improving working conditions as a means of improving worker health and wellbeing, including fertility.