Residential mobility in a cohort of primiparous women during pregnancy and post-partum
Dr Camille Raynes-Greenow, level 4, Wallace Freeborn Building, North Shore Hospital, St Leonards, New South Wales 2065. Fax: (02) 9906 6742; e-mail: firstname.lastname@example.org
Objective: To quantify the residential mobility rate in a population of pregnant women expecting their first baby.
Method: We verified residential mobility in a cohort of 585 primiparous Australian women who were enrolled in late pregnancy and had participated in a randomised controlled trial and followed-up to at least 16 weeks post-partum.
Results: We found a residential mobility rate of 19%. Movers and non-movers differed by socio-demographic factors, with movers more likely be younger, relative risk (RR)=2.14 (95% confidence interval (CI) 1.41-3.13), and not living with a partner RR=2.46 (95% CI 1.60-3.77).
Conclusion: Most prospective epidemiological studies can expect some attrition in the study population. The family formation period is acknowledged as a highly mobile time and this mobility may contribute to loss to follow-up.
Implications: Researchers planning prospective studies in pregnant populations should consider the impact of residential mobility, especially differential mobility, and implement strategies to reduce attrition and optimise response rates.
This study aimed to quantify residential mobility in a cohort of primiparous (pregnant expecting their first baby) women. Retention of participants in prospective epidemiological studies is an important consideration for researchers. Most epidemiological studies can expect some attrition in the study population. There are two underlying concerns with attrition or loss to follow-up: first, that losses may be differential and related to either the intervention/exposure and/or the outcome; and, second, significant loss to follow-up could reduce the sample size.1 These are serious concerns as they may introduce bias into the results and the latter could reduce the study's power. For pragmatic reasons, postal questionnaires are frequently used for follow-up. There has been considerable research into methods that may increase response rates to postal questionnaires. A Cochrane Systematic review of methods for increasing response rates to postal questionnaires found 372 eligible trials that included 98 different methods.1 Yet one of the fundamental requirements to ensure optimal response rates is complete follow-up. For most prospective studies, recording of participants’ correct residential address is essential. However, the impact of residential mobility on follow-up has not been considered.
In Australia, the most recent data available from the 2006 National Census found that 41% of the population changed their place of residence sometime in the past five years and people aged between 25-34 years were the most mobile age group.2 Residential mobility is known to be high in the family formation and childbearing years.3 These Australian patterns are consistent internationally and are replicated in the United States and Europe.3,4 Despite knowing that the childbearing period is associated with high mobility, there has been limited assessment of the incidence of residential mobility by researchers in the perinatal field. In other health areas, a paper from 1969 investigated residential mobility and the effect on prospective studies of chronic illness5 and a more recent paper examined change in cancer risk associated with residential mobility.6 In the perinatal field, one paper studied exposure misclassification for adverse pregnancy outcomes due to residential mobility during pregnancy.7 For prospective studies, the impact of residential mobility both during pregnancy and after delivery has not been considered, although the childbearing population is assumed to be highly mobile. This study aimed to quantify residential mobility in a cohort of primiparous women.
Place, time and study type
This cohort study was conducted post-hoc and utilised the data from an obstetric randomised controlled trial of 585 primiparous women. The protocol for this trial has been published elsewhere8 and the manuscript of results is in preparation. The trial was conducted in inner city Sydney, Australia, in two obstetric hospitals between September 2004 and April 2006. One of the two participating hospitals was public and women could choose between the midwife-led antenatal clinic, the midwife-run birth centre clinic or use ‘shared care’ (joint antenatal care between either antenatal clinics or their own general practitioner who is registered with the hospital). The other participating hospital was private and all participants received private care from their chosen obstetrician.
Participants were enrolled in person in late pregnancy at approximately 37 weeks’ gestation. They completed a questionnaire that collected socio-demographic information such as age, marital status, residential address, education, gestation, type of antenatal care, language spoken at home, and alternate contact (parent) and mobile telephone number. All participants were followed-up approximately one week later. This first follow-up was usually conducted face-to-face; however, for many participants it was by phone, and residential address was verified. The second follow-up was conducted approximately 12-16 weeks post-partum and involved a brief pre-notification telephone call immediately followed by a mailed questionnaire, with residential address confirmed during the telephone conversation. Subsequent contact was made with a proportion of women (∼20%) who were invited to participate in a qualitative interview, the timing of this contact varied between 16 weeks to 12 months post-partum and residential address was once again confirmed.
Any attempted or achieved contact with participants was noted in the trial log book. Details of change of address or contact information was recorded in the log book, which was the main source of data for this study.
Verification of mobility status
Women who had any of the following indications were classified as a ‘mover’: those we received verbal or written notification of change of address, or residential telephone disconnection, or our mail was returned as ‘unknown at this address, return to sender’. A very small number of change of addresses (n=3) were identified from medical records and these were used as notification of residential mobility. We assumed responders to the mailed questionnaire that we could not verbally confirm their residential address with as unchanged residence and women whose address was confirmed as unchanged. Mobility status of non-responders to the mailed questionnaire and phone calls (including no answer to phone calls or no message facility, mobile phone turned off or no longer in operation), for which we could not confirm residential address were classified as ‘unknown mobility status’.
This study had ethical approval from the Human Ethics Committees of the participating hospitals, and one of the author's (CRG) own institution. Participants’ details and data were kept separate and in a secure environment at all times as per ethical requirements.
Summary statistics were calculated using frequency and contingency table analyses. Estimates of relative risks and 95% confidence intervals (95% CI) were calculated using Epinfo 20009 to compare socio-demographic characteristics (age, highest level of education, living with partner or not, language spoken at home and type of antenatal care) between movers and non-movers. Sensitivity analyses were performed by reclassifying women with unknown mobility status as movers and then as non-movers to assess the extent of the relationship between response rate and residential mobility.
Of the 585 pregnant women who participated in the RCT, 109 (19%) were classified as mobile sometime during pregnancy and/or up to 12 months post-partum (95% CI 15-21%). There was no record of women changing residence more than once during the study period. Mobility status was considered unknown for 37 (6.3%) women as it was unknown whether these women were non-responders at an unchanged address or if they had moved.
Table 1 presents the maternal characteristics and mobility status of participants. The mean age of participants was 31 years. For 57%, the highest level of education was a university degree, 72% spoke English at home, and 88% were living with a partner. Compared with women aged 25-30 years, women classified as mobile were more than twice as likely to be younger, (<25 years), RR=2.14, (95% CI 1.41-3.13) and women greater than 35 years of age were the least likely to move, RR=0.36 (95% CI 0.15-0.80). Women not living with a partner were twice as likely to have changed residence than women who reported that they were co-habiting, RR=2.46 (95% CI 1.60-3.77). Compared with women who had a higher level of education, women with less education had a greater tendency to move, however this result was not statistically significant, RR=1.34 (95% CI 0.87-2.02). Women receiving shared GP antenatal care were more than 50% less likely to move RR=0.59 (95% CI 0.32-1.07). Compared with movers, a higher proportion of participants with unknown mobility status were older (df=3, χ2=11.0022, p=0.0117) and received GP-shared care (df=3, χ2=16.2281, p<0.001), but they did not differ in regards to level of education, partner status or language spoken at home.
Table 1. Characteristics of participants by moving status.
|Age (years)|| || || || || || || |
| 18-24||25||(43.1)||30||(51.72)||2.14 (1.41-3.13)||3||(5.17)|
| 25-30||40||(19.6)||145||(72.9)||1.00 referent||15||(7.50)a|
| 31-35||38||(15.77)||192||(79.67)||0.76 (0.51-1.13)||11||(4.56)|
| 36||6||(6.98)||72||(83.72)||0.36 (0.15-0.80)||8||(9.30)|
|Education|| || || || || || || |
| Secondary||23||(23.96)||67||(69.79)||1.34 (0.87-2.02)||6||(6.25)|
| Technical/other||27||(17.20)||122||(77.71)||0.96 (0.67-1.35)||8||(5.10)|
| University||59||(17.77)||250||(75.30)||1.00 referent||23||(6.93)|
|Partner|| || || || || || || |
| Not living with partner||14||(42.42)||17||(51.51)||2.46 (1.60-3.77)||2||(6.06)|
| Living with partner||95||(17.2)||422||(76.4)||1.00 referent||35||(6.34)|
|Type of pregnancy careb|| || || || || || || |
| Public||41||(18.98)||166||(76.85)||1.00 referent||9||(4.17)c|
| GP shared||12||(10.26)||91||(77.78)||0.59 (0.32-1.07)||14||(11.97)|
| Birth centre||44||(23.66)||128||(68.82)||1.29 (0.88-1.87)||14||(7.53)|
| Private||12||(18.18)||54||(81.82)||0.92 (0.51-1.64)||0|| |
|Language|| || || || || || || |
| English at home||82||(19.29)||319||(75.06)||1.00 referent||24||(5.65)|
| Other language||27||(16.88)||120||(75.0)||0.91 (0.61-1.34)||13||(8.13)|
In the original trial of this study there were no differences in response rate between the control and the intervention arm, RR=1.07 (95% CI 0.78-1.48), or any differences between the trial arms and mobility rates, RR=1.38 (95% CI 0.94-2.03). However, a slightly larger proportion of women who moved (28%) did not reply to the second follow-up questionnaire compared with those women who did not move (22%), RR=1.27 (95% CI 0.89-1.8).
We explored the possibility of non-response due to mobility using sensitivity analyses. We reclassified all women with ‘unknown mobility status’ (n=37) as movers and the risk of non-response due to residential mobility more than doubled, RR=2.11 (95% CI 1.65-2.71). We then reclassified all women with ‘unknown mobility status’ as non-movers (and responders), yielding a RR=1.38 (95% CI 0.97-1.96), which is consistent with the trend, although not statistically significant.
There was no record of any trial participant moving out of metropolitan Sydney, interstate, or leaving Australia. The only data we have of this occurring are from the non-participant group of 114 women who refused the trial invitation, of which seven (6%) gave their reason as ‘moving out of area’, including two who were ‘leaving Australia’.
Although methods for increasing response rates have been well investigated, there have been limited studies on the effect of residential mobility on attrition in prospective studies. This is one of the only studies to measure residential mobility in a cohort of women during pregnancy and the post-partum period. Our result of 19% of primiparous women changing residence some time during pregnancy and up to 12 months post partum is a useful estimate of mobility. The only other similar study found a rate of 12%; however, it was in a smaller population and limited to pregnancy.7
Importantly, we found that residential mobility varied by socio-demographic factors. Being younger and not having a partner were associated with mobility and these variables have been found to be associated with lower socio-economic status and adverse pregnancy outcomes.10 In contrast, women who were older and with higher education were the least likely to change residence. These results are consistent with findings from Bright's study, which showed lower socio-economic groups changed residence most frequently.5 Correspondingly, response rates to postal questionnaires are also known to be poorer in lower socio-economic groups.11 We propose that residential mobility may account for a proportion of the non-response rates associated with postal questionnaires, and methods to increase response rates may be as simple as maintaining the correct residential address details of study participants. In the women where we could not determine their mobility status, we found that they shared similar characteristics to movers including education level, partner status and language spoken at home, although they did differ with regards to age and type of antenatal care.
In our original trial the rate of mobility and loss to follow-up was non-differential between trial arms, implying that loss to follow-up was not related to the intervention. An interesting result was the trend to non-response after having moved, and although it was not significant, it highlights a potentially large problem. To alleviate this problem the timing of contact and follow-up should be considered. Our follow-up was approximately 19 weeks (less than five months) since the last contact, and may be short enough to catch most movers. We were unable to test this as we did not record the date that we received notification of ‘change of address’. However, decreasing the time between follow-ups or building in a properly evaluated system to remind participants to notify ‘change of address’ may be useful to capture mobile study participants.
Although ascertainment of mobility status in this study was high (94%), we hypothesise that the 19% mobility rate is likely to be an under-estimate of the true mobility rate for several reasons. First, our findings highlight a trend towards non-response by participants after having moved and results of the sensitivity analyses suggests that if the unknown group (6%) were all mobile, the rate of mobility could be as high as 25%. Second, 6% of women who declined to participate in the trial stated ‘moving’ as their reason for non-participation in the original trial. This suggests a possible participation bias by women with an imminent residential move, and thus may contribute to an under-numeration of mobility. Third, university-educated women were over-represented in our sample and women in this study are slightly older than both State (29.9 years) and national mothers (29.8 years).12 We found that women with these characteristics were less likely to move. Finally, our data were limited to primiparous women and did not allow us to analyse the effect of time and increased parity. We estimate that the overall residential mobility rate is likely to increase with time and with the addition of another child, based on anecdotal data that suggests families expecting a second child report moving out of the inner city area (with typically smaller house sizes) more frequently.
Our results highlight that strategies to minimise the impact of residential mobility may be beneficial in reducing loss to follow-up. A potential solution, although not new, is to increase the sample size of the population to ensure the sample is adequately powered regardless of loss. Over-sampling within certain strata (such as lower socio-economic groups) may also be required as our results showed differential mobility by these groups, which is a major source of bias. A second, more cost-effective strategy would be to collect additional contact details from each participant such as email, multiple phone contacts, and a secondary contact such as someone unlikely to move (e.g. parent) to ensure residential mobility does not affect follow-up. This needs to be carefully considered so as not to unduly burden participants, and evaluated to assess the effectiveness of such methods.
The burden of tracing study participants for new linked data studies when using previously collected data is another problem associated with residential mobility. The likelihood of successful tracing is an important consideration for researchers and ethicists weighing up the difficulties of contacting persons whose data would be used in the study, with the potential harm created by allowing the investigator to proceed without consent.13 Having an estimate of residential mobility may be useful for researchers and ethicists who are considering these issues and estimating this burden for both researchers and participants.
Our study found that residential mobility occurred in at least 19% of a primiparous population. It was more likely to occur among younger, less educated women and women not living with a partner. Strategies to minimise the impact of residential mobility need to be considered to reduce attrition of study participants, and researchers should consider the impact of residential mobility in all study designs.
This research is funded by NHMRC project grant #253635. CRG is an NHMRC Postdoctoral Research Fellow and at the time of this work was supported by an NHMRC Post Graduate Public Health Scholarship. NN is an NHMRC Postdoctoral Research Fellow. CR is an NHMRC Research Fellow. Thanks to Jane Bell for helpful suggestions and Clare Ryan for data collection, and to all the women who generously participated.