Faecal incontinence persisting after childbirth: a 12 year longitudinal study


Correspondence: C MacArthur, Public Health, Epidemiology and Biostatistics, School of Health and Population Sciences, University of Birmingham, Birmingham B15 2TT, UK. Email c.macarthur@bham.ac.uk



To investigate persistent faecal incontinence (FI) 12 years after birth and association with delivery mode history and quality of life.


Twelve-year longitudinal study.


Maternity units in Aberdeen, Birmingham and Dunedin.


Women who returned questionnaires 3 months and 12 years after index birth.


Data on all births over 12 months were obtained from units and women were contacted 3 months, 6 years and 12 years post birth.

Main outcome measure

Persistent FI, defined as reported at 12 years and one or more previous contacts. SF12 assessed quality of life.


Of 7879 women recruited at 3 months, 3763 responded at 12 years, 2944 of whom also responded at 6 years: nonresponders were similar in obstetric factors. Prevalence of persistent FI was 6.0% (227/3763); 43% of 12-year responders who reported FI at 3 months also reported it at 12 years. Women with persistent FI had significantly lower SF12 scores. Compared with only spontaneous vaginal deliveries, women who had one or more forceps delivery were more likely to have persistent FI (odds ratio [OR] 2.08, 95% confidence interval [95% CI] 1.53–2.85) but it was no less likely with exclusively caesarean births (OR 0.93, 95% CI 0.54–1.58). More obese women than normal weight women reported persistent FI (OR 1.52, 95% CI 1.06–2.17).


This longitudinal study has demonstrated persistence of FI many years after birth and shown that one forceps birth increased the likelihood, whereas exclusive caesarean birth showed no association. Obesity, which increased symptom likelihood, is a modifiable risk factor.


Faecal incontinence (FI) is an unpleasant problem that is more common in parous women.[1] The prevalence of FI varies according to definition and method of ascertainment and most studies report rates of postpartum FI after one delivery. In our longitudinal study, prevalence of FI at 3 months after the index birth was 9.6%,[2] however, some FI symptoms did resolve and at 6 years postpartum we found that 59% of women who had FI at 3 months no longer experienced it.[3] Specific aspects of childbirth have been found to be associated with an increased likelihood of FI, including forceps delivery, older maternal age and increasing body mass index[1, 4-6] but again most studies investigate risk factors after one delivery. We have continued our cohort study to a 12-year follow up to enable examination of the effects of delivery history. We have already reported that FI at 12 years was increased if a woman had just one forceps delivery; and was not reduced in women who had all their births by caesarean section.[7] Given that FI does resolve, however, it is also important to examine persisting FI symptoms but few studies have been able to follow a large enough cohort of women for long enough to do this. This paper presents findings on prevalence of persistent FI, effects on quality of life and relationships with delivery mode history and other obstetric and maternal factors. The main research questions were the extent to which postpartum FI persists and whether delivery mode history or other factors are predictive of persistent FI.


All women who delivered over a 12-month period during 1993/94 in three maternity units, in Aberdeen (UK), Birmingham (UK) and Dunedin (New Zealand) were sent a postal questionnaire at 3 months postpartum to assess prevalence of urinary and faecal incontinence. Women with urinary incontinence at this time were eligible to take part in a randomised controlled trial of the effects of an intensive pelvic floor muscle training programme (described elsewhere).[8, 9] Women who responded at 3 months were sent another questionnaire at 6 years postpartum, then another at 12 years, including nonresponders at 6 years but excluding those who requested no further contact in their 6-year questionnaire and known deaths.

Data collection and outcomes

To assess FI, women were asked: ‘do you ever lose control of bowel motions (stool/faeces) from your back passage in between visits to the toilet?’ Response options were ‘never’, ‘occasionally’, ‘sometimes’, ‘most of the time’ and ‘all of the time’. The same wording and response options applied to the question on flatus incontinence—'do you ever lose control of wind (gas) from your back passage in between visits to the toilet?'. The questions were designed by the study team, because at the time of initial recruitment there was no suitable validated questionnaire on FI and we continued with the same form of ascertainment in follow up for consistency.

Subsequent additional questions to women who reported FI were: whether her FI had started before her first birth and whether she ever had passive incontinence (bowel leakage without being aware until it happened). Further questions on FI were about recent use of pads, plugs or constipating medicines to stop leakage: faecal urgency (hold on for 5 minutes when had desire to open bowels); and extent to which FI interfered with everyday life (on a ten-point scale from ‘not at all’ to ‘a great deal’). All women were asked if in the past they had had any surgical treatment or biofeedback/bowel training to treat bowel leakage. The SF12 was also included in the 12-year questionnaire to assess generic health-related quality of life. It is a 12-item tool and has two summary scores, the mental component score (MCS) and the physical component score (PCS), which assess mental and physical functioning respectively, both with a range of 0–100 and a mean of 50.[10]

Faecal incontinence was defined as loss of control of bowel motions occurring with any level of frequency (‘occasionally’, ‘sometimes’, ‘most of the time’ and ‘all of the time’). The main outcome of persistent FI was defined as FI reported at 12 years as well as at least one of the previous contact points of 3 months or 6 years.

A secondary outcome was persistent flatus incontinence. As flatus is known to be a much more common symptom, our definition of persistent flatus was flatus reported as more often than occasional at both 12 years and at least one of the previous contact points of 3 months or 6 years.

Obstetric and maternal data on the index delivery were obtained from the hospital case-notes. Follow-up questionnaires at both 6 and 12 years obtained date and mode of every delivery from each woman to determine delivery mode history. The initial study and both follow ups were approved by ethics committees in each centre.

Research questions and analysis

The main research questions investigated in this paper were the extent to which postpartum FI persists and whether delivery mode history is predictive of persistent FI. Secondary research questions were the extent to which postpartum flatus incontinence that is more than an occasional occurrence persists and whether delivery mode history is predictive of persistent flatus incontinence. The relationship between persistent symptoms and quality of life was another secondary research question.

The variable delivery mode history was created from the reported birth histories at 12 years. This categorised all of a woman's deliveries, including those which predated our first contact with her 3 months following the index birth into: spontaneous vaginal delivery only; caesarean section only; one or more forceps deliveries; one or more vacuum extractions but no forceps; a combination of spontaneous vaginal delivery and caesarean section.

Multiple logistic regression was used to investigate independent associations between delivery mode history and outcomes, adjusting for age at first birth (<25/25–29/30–34/35 + years), total number of births (one/two/three/four or more), current body mass index (BMI; underweight/normal weight/overweight/obese/not known) and ethnic origin (non-Asian/Asian): and to report on other independent associations.

More detailed obstetric factors were only available from case-notes for index births so to investigate further obstetric associations a subgroup analysis restricted to women for whom the index birth was their first (index primiparae) was undertaken. The additional obstetric and maternal variables included were onset of labour (not induced/induced), perineal trauma (intact/episiotomy/laceration), and birthweight (quartiles). Data on degree of laceration were not recorded in a comparable format from all three units so we could not investigate third-degree and fourth-degree tears.

Women with missing values (except for BMI—see above) were omitted from the models.


In all, 10 989 women were sent a questionnaire at 3 months postpartum and 7879 replied. At 6 years, 117 women had requested no further contact and there had been 41 deaths and at 12 years further checking of records identified four additional women from 3 months eligible to receive a 12-year questionnaire. Excluding the 158 women who requested no further contact or who had died, 7725 follow-up questionnaires were sent at 12 years, of which 798 were returned as ‘moved away’ by the post office and 3763 were returned completed by the women, giving a response rate of 48.7% (3763/7725). Among these 3763 women, 2944 (78.2%) replied at all three contact points and 819 (21.8%) at 3 months and 12 years, but not at 6 years. The 2944 responses represent 69.9% (2944/4214) of women who had replied at 6 years. Mean duration of follow up was 12.9 years (SD 0.67; range 10.6–14.3) and mean maternal age at follow up was 42.4 years (SD 4.9; range 27–58).

Comparison of characteristics of respondents (R) and nonrespondents (NR) at 12 years showed some differences (table previously shown in ref. 7). Fewer respondents had been under 25 years at index birth (17.3% for R, 28.5% for NR), fewer were Asian (4.0% for R, 9.1% for NR), fewer were multiparous (53.2% for R, 56.4% for NR) and fewer had an intact perineum (27.0% for R, 36.8% for NR). Delivery mode, onset of labour and length of second stage of index birth were similar. The prevalence of FI at 3 months was lower among respondents (8.2% for R, 9.5% for NR), which is likely to slightly underestimate the prevalence of symptoms at 12 years.

Prevalence, persistence and effects of FI

The prevalence of persistent FI at 12 years was 6.0% (227/3763). Table 1 shows the pattern of FI across the three contact points at 3 months, 6 years and 12 years for all women and separately for women who had had their first baby at our first contact with the cohort at 3 months (index primiparae). Among all women who replied at 12 years, 19.9% (747/3763) had reported FI on at least one of the contact points: total FI prevalence at 12 years was 12.9% (487/3763) compared with 9.7% (287/2944) at 6 years and 8.2% (307/3763) at 3 months. Among the index primiparae the proportions with persistent FI and with FI at the various contact points were the same as for all women in the sample.

Table 1. Faecal incontinence among all women at each contact among those who replied at 12 years (n = 3763) and for primiparae at index delivery (n = 1760)
3 months6 years12 yearsTotal, all womenTotal Index primiparas
n (%) n (%)
  1. a

    Persistent FI.

  2. –, Nonresponse as not all women replied at 6 years as well as 12 years.

Total at 3 monthsTotal at 6 yearsTotal at 12 yearsTotal at any timeTotal at any time
All women All women All women All women  
Index primiparas Index primiparas Index primiparas   Index primiparas
138/1760127/1373213/1760 329/1760
(7.8%)(9.2%)(12.1%) (18.7%)

Resolution and new onset of FI symptoms both occurred. Among the 307 women who had reported FI at 3 months, 131 (42.7%) also reported it at 12 years whereas in 176 (57.3%) symptoms had resolved by 12 years. Even among the latter 19.9% (35/176) had still had FI at our 6-year contact so it had not just been a short-term problem. Of the 3456 women with no FI at 3 months there were 356 (10.3%) who reported it at 12 years, and in 96 (27.0%) of these women it had persisted from 6 years. Again, the pattern was the same among those who had given birth for the first time at our 3-month contact (index primparae).

We also calculated the proportion of new cases of FI at each contact point among the women who responded at all three times (n = 2944): 8.4% (244/2944) at 3 months, 6.1% (180/2944) at 6 years and 5.9% (173/2944) at 12 years. Some of the multiparae at the index birth may have had postpartum FI after a previous birth so we also calculated these rates separately for the index primiparae. The proportion of new cases of FI in this group was 7.9% (108/1359) at 3 months, 6.0% (81/1359) at 6 years and 5.7% (78/1359) at 12 years.

Most of the 227 women who reported persistent FI reported it as occurring occasionally at 12 years, with 27.3% (62/227) reporting it occurring more frequently than this. Very few of the women with persistent FI (7.9%, 18/227) said that it had started before having their first baby. There was indication of some severe symptoms: among the women with persistent FI 22.9% (52/227) said at 12 years that they sometimes had leakage without being aware until it had happened; 7.5% (17/227) wore a pad or plug and 9.3% (21/227) used constipating medicine to protect against leakage. In terms of interference with everyday life, 14.1% of the women with persistent FI (32/227) reported no effect at all and on a scale of 0–10 where ten is a great deal, 29.1% (66/227) rated it as five or more. Only six women with persistent FI said that they had had surgical treatment for this and three more reported physiotherapeutic treatment.

The quality of life SF12 MCS and PCS were significantly lower (worse) among women with than without persistent FI (Table 2). Given that the group without persistent FI comprised three subgroups (FI at 12 years but not before; FI sometime since index birth but not at 12 years; no FI) we looked separately at mean MCS and PCS scores in these groups (Table 2). It was the group with no FI at all that had the significantly higher SF12 scores although the difference in MCS for those who previously had FI that was now resolved almost reached statistical significance.

Table 2. Quality of Life—comparison of SF12, MCS and PCS scores for faecal incontinence and flatus incontinence
 PersistentNot persistentNot persistent
 At 12 years, not beforeSometime, not at 12 yearsNone
  1. a

    Comparison of persistent FI by FI status.

  2. b

    After removing persistent FI cases.

Faecal incontinence
n (%)225 (6.1)3463 (93.9)255 (6.9)257 (7.0)2951 (80.0)
Diff (95% CI)a 4.5 (3.1–5.9)0.7 (–1.2 to 2.8)1.8 (–0.1 to 3.7)5.0 (3.6–6.5)
P a  <0.0010.4430.069<0.001
Diff (95% CI)a 2.7 (1.3–4.1)−0.3 (–2.2 to 1.6)1.3 (–0.6 to 3.1)3.1 (1.7–4.5)
P a  <0.0010.7390.172<0.001
Flatus incontinence b
n (%)575 (16.6)2888 (83.4)415 (12.0)692 (20.0)1781 (51.4)
Diff (95% CI)a 2.8 (1.8–3.7)−.3 (–1.7 to 1.1)1.9 (0.7–3.0)3.9 (2.9–4.9)
P a  <0.0010.6550.002<0.001
Diff (95% CI)a 1.0 (0.2–1.8)−.4 (–1.5 to 0.8)0.8 (–0.2 to 1.7)1.3 (0.5–2.2)
P a  0.0170.5480.1170.001

We also asked about faecal urgency at 12 years and among the women with persistent FI, 41.0% (93/227) also reported faecal urgency. Among those in the sample without persistent FI but FI at some time, the proportion also reporting faecal urgency was 20.0% (104/520), whereas among those who had no FI at any time only 4.4% (104/2347) reported faecal urgency.

Persistent faecal incontinence and delivery mode history

Persistent FI was significantly more common if any of a woman's births were by forceps, but there was no association for a delivery mode history that included vacuum extraction but no forceps. There was no association between persistent FI and a history of exclusively caesarean section births or where the history comprised spontaneous vaginal deliveries and caesarean sections (Table 3). Current BMI also showed an independent association: obese women had significantly more persistent FI compared with those of normal weight. Significantly more Asian women reported persistent FI, but based on small numbers (n = 10/67). There were trends for more persistent FI with increasing maternal age at first birth and total number of births, but differences were not statistically significant except for age 30–34 years at first birth (Table 3). The comparison shown in Table 3 is between women with and those without persistent FI and the latter includes women with no FI at all as well as those who experienced FI but not persistently. We repeated the logistic regression model excluding women who experienced nonpersistent FI and the same pattern of associations remained (data not shown).

Table 3. Logistic regression of persistent FI and delivery mode history
VariableTotalUnadjusted persistent FI n (%)Adjusted OR(95% CI) P
  1. a

    Only CS subdivisions: only prelabour 7/124; only postlabour 2/109; only pre/postlabour 9/170.

  2. SVD, spontaneous vaginal delivery; CS, caesarean section.

  3. Total n = 3759: four cases excluded from analysis due to missing delivery history data.

Delivery mode history
Only SVD185885 (4.6)Reference 
Only CSa40318 (4.5)0.93(0.54–1.58)0.777
Any forceps95689 (9.3)2.08(1.53–2.85)<0.001
Any vacuum, no forceps24814 (5.6)1.22(0.68–2.19)0.509
SVD + CS29421 (7.1)1.46(0.89–2.40)0.139
Age at first birth, years
≤24127475 (5.9)Reference 
25–29149382 (5.5)1.01(0.72–1.42)0.962
30–3478856 (7.1)1.48(1.00–2.18)0.050
≥3520414 (6.9)1.48(0.79–2.76)0.223
Number of births
One41117 (4.1)Reference 
Two1836110 (6.0)1.50(0.88–2.57)0.135
Three101565 (6.4)1.67(0.94–2.95)0.080
Four or more49735 (7.0)1.67(0.88–3.18)0.119
BMI at 12 years
Underweight615 (8.2)1.71(0.66–4.40)0.267
Normal178697 (5.4)Reference 
Overweight102061 (6.0)1.11(0.80–1.56)0.525
Obese64351 (7.9)1.52(1.06–2.17)0.023
Not known24913 (5.2)0.95(0.52–1.72)0.852
Asian ethnicity
No3598209 (5.8)Reference 
Yes16118 (11.2)2.06(1.20–3.53)0.008

To consider whether other obstetric factors (only available for index births) might be associated with persistent FI, the model was repeated for those for whom the index birth was their first (Table 4). None of these obstetric factors were significantly associated with persistent FI, although there was a nonsignificant increase in women who had a nonintact perineum. The increased likelihood of persistent FI with any forceps birth and lack of effect for only caesarean section deliveries remained with entry of the additional obstetric factors into the model.

Table 4. Logistic regression persistent FI and delivery mode history among index primiparae (n = 1759)
VariableTotalUnadjusted persistent FI, n (%)Adjusted OR(95% CI) P
  1. SVD, spontaneous vaginal delivery; CS, caesarean section.

  2. Total n = 1759: one case excluded from analysis due to missing delivery history data.

Delivery mode history
Only SVD84334 (4.0)Reference 
Only CS2449 (3.7)0.75(0.35–1.62)0.463
Any forceps39541 (10.4)2.65(1.64–4.28)<0.001
Any vacuum, no forceps17112 (7.0)1.78(0.89–3.55)0.101
SVD + CS1067 (6.6)1.45(0.61–3.43)0.403
Age at first birth, years
≤2448722 (4.5)Reference 
25–2966934 (5.1)1.22(0.68–2.17)0.505
30–3446235 (7.6)2.08(1.15–3.75)0.015
≥3514112 (8.5)2.42(1.10–5.30)0.027
Number of births
One40117 (4.2)Reference 
Two92861 (6.6)1.68(0.94–3.00)0.078
Three33721 (6.2)1.57(0.78–3.15)0.205
Four or more934 (4.3)0.78(0.23–2.66)0.695
BMI at 12 years
Underweight293 (10.3)2.28(0.63–8.26)0.209
Normal83542 (5.0)Reference
Overweight48034 (7.1)1.62(1.00–2.62)0.051
Obese30819 (6.2)1.35(0.76–2.40)0.301
Not known1075 (4.7)0.81(0.31–2.16)0.680
Asian ethnicity
No169293 (5.5)Reference
Yes6710 (14.9)4.59(2.08–10.13)<0.001

Persistent flatus incontinence: prevalence and delivery mode history

The prevalence of persistent flatus incontinence (more often than occasional) at 12 years was greater than for persistent FI, being reported by 18.8% of the women (709/3763) (Table 5). Persistent flatus incontinence, experienced with any degree of frequency, was very common, reported by over half of the sample (53.4% 2011/3763).

Table 5. Flatus incontinence among all women at each contact among those who replied at 12 years (n = 3763) and for primiparae at index delivery (n = 1760)
3 months6 years12 yearsTotal all womenTotal Index primiparas
n (%) n (%)
  1. a

    Persistent flatus incontinence.

  2. –, Nonresponse as not all women replied at 6 years as well as 12 years.

Total at 3 monthsTotal at 6 yearsTotal at 12 yearsTotal at any timeTotal at any time
All women All women All women All women  
Index primiparas Index primiparas Index primiparas    Index primiparas
455/1760419/1373551/1760  890/1760
(25.9%)(30.5%)(31.3%)  (50.6%)

The quality of life SF12 MCS and PCS scores for women with persistent nonoccasional flatus incontinence (after excluding women with persistent FI) are shown in Table 2. Overall women with persistent flatus had higher MCS and PCS scores than those with persistent FI but they were significantly lower (worse) than among the women with no persistent flatus.

The pattern of associations with delivery history for persistent flatus (more often than occasional) was the same as for persistent FI: an increase among women who had any forceps birth and no association for vacuum without forceps, for exclusive caesarean births or for mixed spontaneous vaginal deliveries and caesarean sections. There were no significant associations for maternal age at first birth, parity or Asian ethnicity but persistent flatus was more common among women who were overweight and obese at 12 years (Table 6). As for persistent FI, the model including index primiparae and other obstetric factors at the index birth showed no other associations with persistent flatus incontinence (data not shown).

Table 6. Logistic regression of persistent flatus incontinence and delivery mode history
VariableTotalUnadjusted persistent flatus, n (%)Adjusted OR(95% CI) P
  1. SVD, spontaneous vaginal delivery; CS, caesarean section.

  2. Total n = 3759: four excluded from analysis due to missing delivery history data.

Delivery mode history
Only SVD1858318 (17.1)Reference 
Only CS40369 (17.1)0.88(0.66–1.18)0.394
Any forceps956227 (23.7)1.46(1.20–1.77)<0.001
Any vacuum, no forceps24839 (15.7)0.86(0.60–1.24)0.418
SVD + CS29456 (19.0)1.13(0.82–1.55)0.460
Age at first birth, years
≤241274231 (18.1)Reference 
25–291493277 (18.6)1.04(0.85–1.28)0.681
30–34788157 (19.9)1.16(0.91–1.48)0.234
≥3520444 (21.6)1.26(0.86–1.85)0.234
Number of births
One41181 (19.7)Reference 
Two1836360 (19.6)1.01(0.77–1.34)0.919
Three1015170 (16.7)0.84(0.62–1.14)0.264
Four or more49798 (19.7)0.99(0.69–1.42)0.959
BMI at 12 years
Underweight618 (13.1)0.80(0.38–1.71)0.566
Normal1786294 (16.5)Reference 
Overweight1020216 (21.2)1.38(1.13–1.68)0.001
Obese643154 (24.0)1.62(1.29–2.02)<0.001
Not known24937 (14.9)0.89(0.61–1.29)0.529
Asian ethnicity
No3598676 (18.8)Reference 
Yes16133 (20.5)1.14(0.76–1.72)0.516


This large longitudinal study, following women to 12 years after index birth, has shown that 6% had persistent FI from either their 3-month or 6-year follow up. The FI occurring after birth seemed to have become persistent for just under half of these women: of those who had FI when first asked at 3 months, 42.7% also reported it at 12 years. The same pattern of persistence and resolution of FI was evident for women of any parity at the index birth and separately for those who had just delivered their first baby. Although most women with persistent FI reported this as an occasional occurrence, over a quarter said that it was more common than this at the 12-year follow up and just under one-quarter sometimes had leakage without any prior awareness. There was a clear association with quality of life: women with persistent FI had significantly worse SF12 MCS and PCS with the greatest effect on the MCS; yet only nine women reported having surgical or physiotherapeutic treatment. Prevalence of persistent flatus incontinence (more than occasional) was 18.8% and was also significantly associated with reduced quality of life.

We have already reported that delivery mode history was a predictor of any FI at 12 years[7] and this paper has shown the same delivery mode history pattern to be predictive of FI that is persistent. This has not previously been reported in the literature. Relative to having only spontaneous vaginal deliveries, the presence of one or more forceps deliveries in the delivery history was predictive of persistent FI, whereas having delivered exclusively by caesarean section, having one or more vacuum extraction deliveries or a combination of caesarean section and spontaneous vaginal deliveries showed no differences. For any FI at 12 years we had subdivided the exclusive caesarean section group into only prelabour, only postlabour and a mix of both and shown no difference.[7] For persistent FI numbers were too small for us to consider them meaningful, however this issue is of great interest to clinicians so we have reported them in a footnote on Table 2 showing no obvious difference indicated for type of caesarean section. BMI categorised as obese at 12 years was independently associated with persistent FI, as was older age (30–34 years) at first birth and Asian ethnicity. Persistent incontinence of flatus showed a similar pattern of independent associations with delivery history and greater BMI.

There is very little literature on long-term postpartum FI and almost none documenting the extent of persistence. Dolan and Hilton[11] sent postal questionnaires (including the Sheffield pelvic floor questionnaire) to women 20 years after they had had their first baby in one UK unit in 1983–86 and 888 (62%) responded to questions on FI. Prevalence of FI at 20 years was 23.1%, higher than our 12-year prevalence of 12.9%,[7] although only 3% of the women with FI in their study reported it as more often than occasional compared with our 27%. They found that women who had instrumental first births were more likely to have FI at 20 years, consistent with our findings although they did not subdivide type of instrument used. Also consistent with our findings was an association between increasing BMI and 20-year FI. The women in this study had not been contacted before the 20-year questionnaire so there were no data on persistent FI 20 years post birth.

A US population-based study (the Fecal Incontinence Postpartum Research Initiative) sent a postal survey at 3–6 months postpartum (40% response) then a follow up to symptomatic women at 1 and 2 years, producing some data on persistence of postpartum FI.[1, 12] At first contact 29% of the 8774 responders reported anal incontinence (flatus and stool) in the last 30 days and almost half of these included FI.[1] Follow up showed that for all women, anal incontinence prevalence decreased over time from 45% at 1 year to 28% of those initially reporting at 2 years: and with significant effect on several quality of life indicators relative to resolved cases.[12] For those with FI the proportion with persistence after initial contact remained stable up to 2 years, at 35%.[12] This is a little lower than in our study but less than half of the original symptomatic women responded again and the assumption made by the authors was that all nonresponding cases had resolved, which is not likely. They found that women with persistent anal incontinence were significantly more likely to report greater adverse effects on various quality of life items and that this was four to seven times higher for persistent FI. Only 17% had reported their persistent anal incontinence symptoms to a medical provider at 2 years.

Handa et al.[13] in another US study compared three groups of women with: anal sphincter tear; vaginal birth without sphincter tear; and caesarean delivery. They used the SF12 to assess quality of life in women who reported FI at 6 months postpartum. As in the current study, they found both MCS and PCS to be significantly reduced in symptomatic women and there was a greater effect on MCS. Their SF12 scores overall were higher than in our study, unsurprising as they were not investigating persistent symptoms.

Data from longitudinal studies such as this can be analysed in many different ways. It is possible to make more use of the longitudinal nature of the data by, for example, using continence status and delivery history at every measurement point and using either subject-specific or population-averaged longitudinal models for analysis. Using these forms of analysis would answer a different question than the one we wanted to answer. For a start it would answer questions about FI, rather than persistent FI, but would also answer hypotheses about the whole 12 years of the study, rather than how the women were 12 years after recruitment.

Longitudinal studies typically suffer from loss to follow up, which can affect the validity of the findings. Given its duration this study has had many opportunities for missing data. Not all of the cohort responded to the initial questionnaire and there have been varying response rates to follow-up questionnaires. In addition, the questions on FI were answered less frequently than other questions. Missing data were classified into three groups: missing completely at random (MCAR); missing at random (MAR); and missing not at random (MNAR). The first two are called ignorable missing and the last is called nonignorable missing. MCAR means that the missing data are unrelated to both measured and unmeasured data. MAR is unrelated to unmeasured data, but is related to measured data, whereas MNAR is related both to measured and unmeasured data. Models are fairly robust to ignorable missing data, apart from the obvious loss of power. But nonignorable missing data, as is likely the case with this study, is a more difficult issue. The most common method to adjust for missing data is multiple imputation, and propensity scores can also be used. Both of these assume that the data are at most MAR, either overall, or in subgroups defined at baseline. Nonignorable missing data are much more difficult to deal with as they require modelling of the missingness mechanism. This involves using some unverifiable assumptions. It can be done with a sensitivity analysis run in combination with multiple imputation, but software to perform this is rare.

It is likely therefore that the prevalence values from this study are underestimates, as it is clearly about incontinence and that usually means that women with symptoms are more likely to respond. However, when looking at relationships between variables you have to suppose a much more unlikely missingness mechanism to get badly biased results. To take the relationship between FI and mode of delivery as an example you would have to suppose that the response rate for women with the different delivery histories differentially depended on their continence status. Hence, there is likely to be much less of a problem with bias in relationships between variables.

The large size and long duration are the main strengths of the present study. It is the largest postpartum cohort study in the literature with this length of follow up, only small studies having longer follow up.[4, 14] Without long-term follow up it is not possible to ascertain the extent to which postpartum FI, which we now know to be relatively common, is transient or continues as a chronic problem for women and our intention is to follow the sample again at 20 years. Unless a study is large there will be inadequate numbers of women in the various delivery mode history groups to be able to make comparisons of persistent incontinence with an acceptable level of precision.

Obstetric data for the index delivery were available from case records, although data on overall delivery history were obtained directly from the women. Validation studies, however, have shown that women's reports of delivery details are accurate[15] and we checked the accuracy of reports for index birth delivery mode with hospital obstetric records showing that for 98% of caesarean sections, 98% of spontaneous vaginal deliveries, 87% of forceps and 88% of vacuum deliveries data were consistent. Another possible weakness is that a validated questionnaire was not used to assess FI, but at the time of recruitment in 1993 there were none available and we wanted to follow up women using consistent questioning. The BMI was calculated based on women's self-reported height and weight, which is subject to bias, but no other option was available.

The rate of our follow up at 12 years, as already noted, may be considered the main limitation of the study at just under 50% of those initially contacted at 3 months and 54% excluding those known to have moved house. However, it is important to consider the ways in which this might produce bias. The initial intention of the study had been for a single 3-month contact so we did not seek alternative contact details in the event of a house move. At 6 years, alternative contact details to facilitate subsequent follow up were sought and loss between 6 and 12 years was much smaller, with 70% of the 6-year responders responding again. At 12 years, additional methods of tracing were available in the UK centres. As a result, 819 women who had moved by 6 years and so had not responded did so when traced to a different address at 12 years. It is therefore likely that much nonresponse was not purposeful and geographical mobility is not known to be systematically associated with incontinence or delivery mode. We had computerised index birth hospital data for all women and comparisons were made between responders and nonresponders at 12 years to examine possible response bias. The main differences were that fewer younger women (under 25) responded as did fewer from ethnic minority groups, which is a typical response pattern among childbirth populations.[16] Given that the younger women had a lower rate of FI, the age difference is likely to lead to some overestimate from the study of overall prevalence of FI. The reverse was the case for ethnic groups, with Asian women reporting higher rates of FI. Obstetric factors, however, particularly mode of delivery, were broadly similar between responders and nonresponders. So although demographic differences of responders probably have some effect on prevalence estimates, it is much harder to see how they might bias delivery mode history associations.

The findings of this study have some implications for practice. That women who deliver exclusively by caesarean section have no lesser likelihood of persistent FI than those who deliver vaginally is information that medical advisors should pass on to women in their consideration of choice of caesarean section. That just one forceps delivery increases the likelihood of persistent FI might be justification to favour vacuum extraction in the absence of particular indication for forceps but this is probably more controversial. Obesity is a potentially modifiable risk factor. The relationship we have shown with persistent FI, which is consistent with the literature on FI more generally, has clear implications for practice.


This large longitudinal study has demonstrated the persistence of faecal and flatus incontinence many years after birth and shown that one forceps birth increased the likelihood of this, whereas exclusive caesarean birth did not show an effect. Current BMI categorised as obese increased the likelihood, as did Asian ethnicity but based on small numbers. There was a clear association with quality of life yet almost no women reported having treatment.

Disclosure of interest

All authors declare that they have no interests to declare.

Contribution to authorship

CM, CG, DW, RL and PH contributed to the design and analysis of the whole ProLong study. SH, ND and PT-H joined the study group at 12 years and contributed to this part of the study. CM drafted the paper with CG and all authors commented. Other members of the ProLong study group are Adrian Grant and Christine Bain.

Details of ethics approval

Ethical approval from UK centres for the 12-year follow up was obtained from Multicentre Research Ethics Committee Edinburgh, Ref No RG 819/06, November 2007 and from New Zealand National Ethics Committee, Ref No LRS/05/04/009 March 2005.


Grants were obtained from Wellbeing of Women/Royal College of Obstetricians and Gynaecologists for the UK follow up and from the Health Research Council of New Zealand for follow up in New Zealand.


The authors would like to thank the women who took part in the study and Alison McDonald, Anne-Marie Rennie, Jane Cook and Jane Harvey who were involved in assembling the original cohort. Gladys McPherson was responsible for the database design and Margaret MacNeil was involved in the administration of the follow-up study. ■

Commentary on ‘Classes of regression models for longitudinal designs’

Longitudinal designs are powerful designs that can provide valuable insights to understand disease trends or ‘changes in the response’ over time—this design provides an opportunity to infer causal associations. There are two major classes of regression models for longitudinal data: marginal or population-averaged models and cluster-specific or subject-specific regression models (see Zeger et al. Biometrics 1988;44:1049–60 for a review.) Each class of the regression model serves a specific purpose in addressing the scientific objectives of the study.

The population-averaged and subject-specific regression models provide different interpretations of their respective regression parameters. For sake of simplicity, consider a single binary response, Yij, (with 1 indicating those with the outcome and 0 for those without the outcome) and i indexing the subjects and j indexing the times when the outcome is measured on the i-th subject. Let xij denote a single binary exposure measured on the i-th individual at the j-th time. In the setting of a cross-sectional study, a simple logistic regression model to evaluate the exposure–outcome relationship (at a single time-point) typically yields an odds ratio that is interpreted as the ‘ratio of the odds of the response among those with the exposure versus the odds of the response among those without the exposure’. Stated differently, the odds ratio contrasts the two odds between the (sub)populations based on exposure status—hence these models are referred to as population-averaged models.

Measuring the outcome repeatedly over time on the same subject (as is the defining feature in a longitudinal design) induces an inherent ‘dependency’ amongst the responses over time. Repeated observations on the same subject over time are likely to be more similar to one another than the same number of observations on different subjects; this is called ‘dependency’. As an example, measuring the systolic blood pressure at time t1 in an individual will be statistically correlated with the blood pressure measurement at time t0 in the same individual (and hence correlated across time.) Failure to adjust for this dependency, also referred to as the ‘intra-cluster correlation’, will lead to imprecise estimation of the variance of the regression coefficients in a model that regresses Yij on xij; hence the variance of the odds ratio and the associated 95% confidence interval (and the associated P-value) will be biased, although the parameter estimates themselves will be approximately unbiased (in large samples of, say, 50 subjects or more). Liang and Zeger (Biometrika 1986;73:13–22; Biometrics 1986;42:121–30) proposed the generalised estimating equations, or GEE, method to estimate model parameters having population-averaged interpretations similar to cross-sectional studies for independent responses. In the regression model analysis based on GEE, the intra-subject dependency is treated as a ‘nuisance parameter’ for which adjustments to standard errors of estimated regression coefficients are made. Further details of the methods outlined above and an application in perinatal epidemiology can be found in the paper by Ananth et al. (Ann Epidemiol 2005;15:293–301.)

The subject-specific regression model, on the other hand, was developed for a purpose that is different from that of the marginal regression models. When the objective of the study is to specifically estimate parameters with subject-specific interpretations, subject-specific regression models with random effects may be relevant, particularly for modifiable risk factors. The odds ratio derived from a subject-specific logistic regression model is defined as the ‘odds of the response for a subject exposed to a risk factor, divided by the odds for the same subject unexposed to the risk factor’. (Diggle et al. Analysis of Longitudinal Data 2nd edn, New York: Oxford University Press; 2002.) For example, if one is interested in estimating the association between maternal smoking (a binary exposure) and the risk of perinatal death, the odds ratio for the smoking variable derived from a random-effects logistic regression model would be interpreted as the ‘ratio of the odds of perinatal death for a smoking mother to the odds of perinatal death had the same mother been a non-smoker or if she quit smoking’. The odds ratio contrasts the two odds of perinatal death for ‘change in smoking status’ for an individual—hence these models are referred to as subject-specific regression models.

Another important issue when analysing with repeated measures in the setting of a longitudinal design is missing data. Although there are different patterns of missing data, it will suffice to note here that for valid inferences from GEE for marginal regression models, the pattern of missing data must follow the principle of responses ‘missing completely at random’ or MCAR; this terminology refers to a subject whose probability of a response being missing does not depend on its possibly unobserved value nor on the previous responses of the subject in a longitudinal study. This is a more stringent criterion than the assumption of ‘missing at random’, or MAR, for the subject-specific models estimated by the method of maximum likelihood where dropout may depend on previously observed responses but not on the current unobserved response. A complete treatment of missing data in the setting of longitudinal designs is beyond the scope of this commentary, but can be found in the classical text book by Diggle et al. (Analysis of Longitudinal Data 2nd edn, New York: Oxford University Press; 2002.)

In this issue of BJOG, MacArthur et al. examine the risks of faecal incontinence (FI) and quality of life 12 years after childbirth in a cohort of 3763 women with follow up of 12 years. The primary objective was to estimate the prevalence of FI based on the mode of delivery (caesarean section versus vaginal delivery) and quality of life (QoL) at 3 months, and at 6 and 12 years after delivery. A secondary objective was to estimate the extent to which FI was associated with declining QoL. As both objectives of the study refer to the estimation of average prevalence of FI and QoL, and to estimate and contrast the risks of FI among women who underwent caesarean delivery versus those who underwent vaginal deliveries (the two subpopulations), the marginal regression model is most appropriate to address the research objectives of the study.

Disclosure of interests

Dr Ananth serves as the statistical editor of BJOG. He is also the Editor-in-Chief of Paediatric and Perinatal Epidemiology, an international journal published by Wiley-Blackwell.

  • CV Ananth,a JS Preisserb

  • aDepartment of Obstetrics and Gynecology, College of Physicians and Surgeons, Columbia University,New York, NY, USA

  • bDepartment of Biostatistics, UNC Gillings School of Global Public Health, Chapel Hill, NC, USA