February 9, 2017

shift work linked to lowered fertility in women

Publishing in Occupational & Environmental Medicine new research claims women working non-daytime shifts and those with physically demanding jobs had fewer mature oocytes – on average – compared with women who worked day-only shifts.

Roundup comments accompanied this analysis.

Title, Date of Publication & Journal

Occupational factors and markers of ovarian reserve and response among women at a fertility centre

Published: Tuesday 7th February 2017

BMJ Journal: Occupational & Environmental Medicine

Study’s main claims – and are they supported by the data

The paper’s main claims involve novel measures of fecundity in a study of women undergoing assisted reproductive technology (ART) in United States of America (USA).

The paper does not present evidence that these measures are directly linked with physically demanding jobs and shift work in the general population (neither the USA general population or extrapolating to the UK population). Nor does the data show a direct link between these measures of fecundity and lowered fertility in women. (Neither within women in the general population, nor even within the study itself).

The press release title “Physically demanding jobs and shiftwork linked to lowered fertility in women” is not supported by the data, nor the paper as written.

Strengths/Limitations

Study population and novel measures

The key limitation is stated by the authors, “it may not be possible to generalise our findings to couples conceiving without medical intervention”. The paper does not support claims about the women in the general population.

The reason the paper cannot talk about the wider population is linked to a key interests of the paper, novel measures of fecundity. As the authors state, “… directly observe many bio-markers of fecundity which cannot be observed in couples attempting to conceive naturally”.

Reporting significant and non-significant results together

The press release mentions “reductions of nearly 9% and nearly 14.5%” (page 4 of the paper, written as 8.8% and 14.4% respectively). However, the 9% reduction is not statistically significant, with a p-value of 0.08. Hence it should not be reported alongside the 14.5%.

(Possible typo: “In regard to markers…” (p4) gives a p-value of 0.007 whereas the Table 3 gives 0.004 for 14.4% reduction. The p-value of 0.007 relates to the unadjusted comparison. Mixing unadjusted and adjusted results?)

Different treatments within the studying

There are three different ovarian stimulation protocols within the IVF part of the study (page 2). We are not given any details of how many women underwent each protocol nor whether this was accounted for in the analysis.

Types of women in the studying

There seems to be an assumption in the classification that all the women are working. This is not mentioned as an entry requirement into the study.

Again, although not explicit, the paper says “enrolled in a prospective cohort study of couples presenting to an academic fertility centre”, implying all enrolled women where in a couple (although further details are not given: married, co-habit, same-sex).

Arbitrary age-split

Splitting people into young/old groups is common in epidemiological analyses, but usually the groups are aligned to regular intervals (10-19, 20-29, etc) or accepted age milestones (legal adult, retirement, etc).

It is not clear why “older women” are defined as 37 and above. No where does the paper justify the choice of 37 years old (the median age of 35 would have had some justification).  So we are left to wonder how robust the conclusions are to picking, for example 35 years and up as the split.

Grouping response levels

The three occupation exposures are grouped into binary splits.  For very small categories, grouping is standard practice. For work shifts, only 42 women were in the non-day categories, so it is sensible to merge into ‘day’ versus ‘not-day’ shift. However it is less clear why the ‘sometimes’ and ‘often’ moving or lifting categories were merged.

Without being pre-arranged before seeing the data, it raises a potential suspicion that the significant results are not robust to analysing the responses separately (analyse different groupings until one analysis gives a significant result).

Changing response groupings for ‘lifting or moving’

The discussion (and press release) state: “Taken together with our results, it appears that lower oocyte quality could be one pathway mediating the relationship between high frequency of moving or lifting heavy loads at work and reduced fecundity”.

The paper does not analyse “high frequency” lifting or moving, it compared ‘never’ with ‘sometimes/often’ – a mixed category.

Not a longitudinal study

The occupational-exposures were taken as the average over the previous year.  So there is no evidence on the short/long-term impact of these occupational exposures.

The age range in the study is substantial, 18-45 (we are not given the range of ages of participants, only the IQR which is 32-38). Short-term impacts will affect this age range differently.

Confounding

As mentioned in the paper, there is an association between education and whether occupation includes physical exertion or heavy lifting. This may indicate a more complex pattern of exposures, lifestyle and background – e.g. social class or household income. However, this paper is on a relatively small sample (less than 500), so we cannot expect it to answer all these questions. It is a first step to investigating these issues, and should be judged as such.

Glossary 

(IQR) Inter-quartile-range: range of the middle 50% of the data

Confounding: Characteristics/traits that explain an apparent relationship

Fecundity: Biologic capacity for reproduction

(Measures of fecundity are distinct from fertility)

Fertility: Demonstrated fecundity, usually measured by live births

Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.