SEARCH

SEARCH BY CITATION

Keywords:

  • Cohort study;
  • recurrent anal sphincter rupture;
  • risk factors

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

Please cite this paper as: Jangö H, Langhoff-Roos J, Rosthøj S, Sakse A. Risk factors of recurrent anal sphincter ruptures: a population-based cohort study. BJOG 2012;119:1640–1647.

Objective  To determine the incidence and risk factors of recurrent anal sphincter rupture (ASR).

Design  Population-based retrospective cohort study.

Setting  Data were taken from the National Medical Birth Registry, Denmark.

Population  Patients with a first and a second vaginal delivery in the time period 1997–2010.

Methods  Univariate analysis and multivariate logistic regression were used to determine risk factors of recurrent ASR.

Main outcome measures  The incidence of recurrent ASR and odds ratios for possible risk factors of recurrent ASR: age, body mass index, grade of ASR, birthweight, head circumference, gestational age, presentation, induction of labour, oxytocin augmentation, epidural, episiotomy, vacuum extraction, forceps, shoulder dystocia, delivery interval and year of second delivery.

Results  Out of 159 446 women, 7336 (4.6%) experienced an ASR at first delivery, and 521 (7.1%) had a recurrent ASR (OR 5.91). The risk factors of recurrent ASR in the multivariate analysis were: birthweight (adjusted OR, aOR, 2.94 per increasing kg, 95% CI 2.31–3.75); vacuum extraction (aOR 2.96, 95% CI 2.03–4.31); shoulder dystocia (aOR 1.98, 95% CI 1.11–3.54); delivery interval (aOR 1.08 by year, 95% CI 1.02–1.15); year of second delivery (aOR 1.06, 95% CI 1.03–1.09); and prior fourth-degree ASR (aOR 1.72, 95% CI 1.28–2.29). Head circumference was a protective factor (aOR 0.91 per increasing cm, 95% CI 0.85–0.98).

Conclusions  The incidence of recurrent ASR was 7.1%. Risk factors of recurrent ASR were excessive birthweight, vacuum extraction, shoulder dystocia, delivery interval, year of second delivery and prior fourth-degree ASR. A larger head circumference reduced the risk of recurrent ASR.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

Anal sphincter rupture (ASR) is a serious complication that affects 5% of nulliparous women with vaginal delivery in Denmark.1 ASR is frequently associated with anal incontinence (involuntary loss of flatus, liquid or solid stool),2,3 which has a negative impact on quality of life.4 Furthermore, one-third of patients with a previous ASR do not deliver subsequently.5 The two-thirds who become pregnant are at higher risk of recurrent ASR (OR 2.5–7.7),5–11 and after subsequent vaginal delivery their risk of long-term anal incontinence increases from 25 to 62%.3

Instrumental delivery,6,8,9,12 midline episiotomy7,8 and excessive birthweight5,6,11 have been identified as the main risk factors for recurrent ASR. The importance of other possible risk factors, such as grade of ASR in first delivery, head circumference, gestational age, presentation, induction of labour, oxytocin augmentation, lateral episiotomy and shoulder dystocia, have not previously been reported in larger population-based settings, and therefore need further investigation. Knowledge about the prevalence of recurrent ASR and a better understanding of obstetrical risk factors associated with vaginal delivery after previous ASR would enable us to improve counselling and decision-making regarding mode of delivery.

The aim of this population-based study was to determine the incidence and risk factors of recurrent ASR.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

The study is a population-based cohort study. Data were retrieved from the Medical Birth Registry (MBR), National Board of Health, Denmark. Patients with a first and a second vaginal delivery in the time period 1997–2010 (n = 159 446) were included. Patients with preterm deliveries and multiple deliveries were excluded.

An ASR was classified according to the Royal College of Obstetricians and Gynaecologists (RCOG) classification.13 A third-degree ASR was defined as a partial or complete disruption of the anal sphincter muscles, which may involve only the external anal sphincter (EAS) or both the EAS and the internal anal sphincter (IAS) (grade 3a, <50% of EAS; grade 3b, >50% of EAS; grade 3c, both EAS and IAS). A fourth-degree ASR was defined as a disruption of the anal sphincter muscles in combination with a tear of the rectal mucosa. ASR was identified by the International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD 10), codes O70.2 and O70.3, which have been validated in the Danish MBR by comparison with medical records.14

We had information on maternal age (at the time of second delivery), maternal body mass index (BMI, reported prior to the second pregnancy, but only from 2004 onwards), date of first and second delivery (from which delivery interval and calendar year of second delivery are known), grade of ASR (grades 3a and 3b/c were reported separately from 2002, and reported as ‘unspecified third-degree ASR’ before 2002; for further analysis third-degree ASRs were combined), birthweight, head circumference, gestational age, presentation, induction of labour, oxytocin augmentation, epidural, mediolateral episiotomy, vacuum extraction, forceps and shoulder dystocia.

To compare the distribution of risk factors between patients with and without recurrent ASR, the Kruskal–Wallis test was applied for continuous risk factors, and Pearson’s chi-square or Fisher’s exact tests were applied for categorical risk factors, with Fisher’s exact test being applied only when the expected cell counts were below five. Spearman’s rank correlation was used to determine the correlation between continuous variables.

Univariate and multivariate logistic regression analyses were performed to determine the associations between recurrent ASR and risk factors. Odds ratios (ORs) and the corresponding 95% confidence intervals (95% CIs) were calculated. For all models, the linearity of the continuous risk factors was assessed, and as none of these demonstrated departures from linearity they were all included in their original form in the analyses. In the multivariate analysis, the presence of possible interactions between the risk factors was investigated, but no such interactions were found. The multivariate logistic regression model was reduced to determine the set of risk factors being associated with the risk of recurrent ASR. A stepwise backward elimination and forward addition procedure was used to reduce the model. All tests were based on the maximum-likelihood principle, with P < 0.05 considered to be significant.

The adequacy of the reported multivariate model was evaluated using several techniques. A shrinkage factor was estimated to determine the need for shrinkage of regression parameter estimates (adjustment for overfitting).15,16 Model calibration was further assessed with the Hosmer–Lemeshow goodness-of-fit test. The discriminative ability of the selected model was assessed by the area under the receiver operating characteristic curve (AUC), with bootstrap optimism correction based on 1000 bootstrap samples.17,18

Statistical analyses were performed using sas 9.2 (SAS Institute, Cary, NC, USA) (SAS Institute Inc., Cary, NC, USA). The study was approved by the Danish Data Protection Agency.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

A total of 159 446 patients met the inclusion criteria. Of the 7336 (4.6%) patients with a third- or fourth-degree ASR at first delivery, 521 (7.1%, 95% CI 6.5–7.7%) had a recurrent ASR (OR 5.91, 95% CI 5.35–6.53, P < 0.0001). Table 1 shows the distribution of potential risk factors of recurrent ASR. In patients with a prior fourth-degree ASR, 46% (95% CI 43–49%, n = 566) had a subsequent vaginal delivery, and of those 60 (11%, 95% CI 8–13%) had a recurrent ASR.

Table 1. Potential risk factors of recurrent anal sphincter rupture (ASR) in patients with an ASR in their first delivery (n = 7336)
 Recurrent anal sphincter rupture n = 521No recurrent anal sphincter rupture n = 6815 P
  1. *BMI was reported in 4360 patients.

  2. **Birthweight was reported in 7316 patients.

  3. ***Head circumference was reported in 7234 patients.

  4. ****Oxytocin augmentation was reported in 7292 patients.

  5. *****From Fisher’s exact test.

Maternal factors
Maternal age (years)
 Median (IQR)31 (29–33)30 (28–33)0.0076
 <2517 (3.3%)368 (5.4%)0.0349
 25–30157 (30.1%)2290 (33.6%)
 30–35267 (51.3%)3158 (46.3%)
 >3580 (15.4%)999 (14.7%)
Pre-pregnancy BMI (kg/m2)*
 Median (IQR)23.9 (21.5–27.2)23.3 (21.0–26.2)0.0090
 BMI < 2041 (11.6%)572 (14.3%)0.0307
 BMI 20–25168 (47.6%)2083 (52.0%)
 BMI 25–30101 (28.6%)888 (22.2%)
 BMI > 3043 (12.2%)464 (11.6%)
Delivery interval
 Median (IQR)2.9 (2.3–3.7)2.7 (2.1–3.5)0.0004
Calendar year of second delivery
 Median (IQR)2006 (2003–2008)2005 (2002–2008)<0.0001
Grade of ASR at first delivery
 Third degree461 (88.5%)6309 (92.6%)0.0007
 Fourth degree60 (11.5%)506 (7.4%)
Fetal factors
Birthweight (kg)**
 Median (IQR)3.93 (3.60–4.22)3.73 (3.43–4.04)<0.0001
 <3.00011 (2.1%)310 (4.6%)<0.0001
 3.000–3.50071 (13.6%)1691 (24.9%)
 3.500–4.000202 (38.8%)2846 (41.9%)
 4.000–4.500173 (33.2%)1603 (23.6%)
 >4.50064 (12.3%)345 (5.1%)
Head circumference (cm)***
 Median (IQR)36 (35–37)35 (35–36)0.0008
 <337 (1.4%)165 (2.5%)0.0138
 33–34105 (20.4%)1507 (22.4%)
 35–36261 (50.6%)3562 (53.0%)
 >36143 (27.71%)1484 (22.1%)
Gestational age (days)
 Median (IQR)283 (278–288)282 (277–287)0.0007
Presentation
 Occiput anterior479 (91.9%)6441 (94.5%)0.0040
 Occiput posterior30 (5.8%)196 (2.9%)
 Other11 (2.1%)160 (2.4%)
Obstetrical factors
Induction of labour81 (15.6%)855 (12.6%)0.0478
Oxytocin augmentation****66 (12.7%)600 (8.9%)0.0035
Epidural39 (7.5%)453 (6.7%)0.4608
Episiotomy43 (8.3%)636 (9.3%)0.4116
Vacuum extraction42 (8.1%)175 (2.6%)<0.0001
Forceps1 (0.19%)1 (0.01%)0.1370*****
Shoulder dystocia17 (3.3%)61 (0.9%)<0.0001

Risk factors of recurrent ASR from univariate logistic regressions are presented in Table 2, which show that maternal age, delivery interval, calendar year of second delivery, fourth-degree ASR in first delivery, birthweight, head circumference, gestational age, occiput posterior presentation, oxytocin augmentation, vacuum extraction and shoulder dystocia are significant risk factors. The full multivariate model (no interactions) included all of the parameters listed in Table 2. In the reduced multivariate analysis (n = 7221; Table 3), delivery interval (adjusted OR, aOR, 1.08 per increasing year, 95% CI 1.02–1.15, = 0.0122), calendar year of second delivery (aOR 1.06, 95% CI 1.03–1.09, < 0.0001), fourth-degree ASR in first delivery (aOR 1.72, 95% CI 1.28–2.29, = 0.0005), birthweight (aOR 2.94 per increasing kg, 95% CI 2.31–3.75, < 0.0001), vacuum extraction (aOR 2.96, 95% CI 2.03–4.31, < 0.0001) and shoulder dystocia (aOR 1.98, 95% CI 1.11–3.54, P = 0.0289) remain significant risk factors. Occiput posterior presentation (aOR 1.73, 95% CI 1.14–2.63, = 0.0488) was a borderline significant risk factor. The estimated shrinkage factor was 0.99 and therefore no shrinkage of the estimated regression parameters was applied. The P value of the Hosmer–Lemeshow test was 0.89, indicating good calibration. The bootstrap optimism-corrected AUC was 0.66 (95% CI 0.62–0.69).

Table 2. Univariate analysis: risk factors of recurrent anal sphincter rupture (ASR)
 OR95% CI P
  1. Crude odds ratios (ORs) based on all possible observations presented with 95% confidence intervals (95% CIs) (n = 7,336).

  2. *BMI was reported in 4360 patients.

  3. **Birthweight was reported in 7316 patients.

  4. ***Head circumference was reported in 7234 patients.

  5. ****Presentation was reported in 7317 patients.

  6. *****Overall P value = 0.0040.

  7. ******Oxytocin augmentation was reported in 7292 patients.

Maternal factors
Maternal age (years)1.031.01–1.060.0061
Maternal BMI before the second pregnancy*1.021.00–1.040.0818
Delivery interval (by year)1.121.06–1.180.0003
Calendar year of second delivery1.061.04–1.09<0.0001
Grade of ASR at first delivery
 Third degree1.0  
 Fourth degree1.621.22–2.160.0015
Fetal factors
Birthweight (kg)**2.522.08–3.05<0.0001
Head circumference (cm)***1.111.05–1.190.0006
Gestational age (days)1.021.01–1.030.0003
Presentation****
 Occiput anterior1.0  
 Occiput posterior2.061.39–3.060.0010*****
 Other presentations0.920.50–1.720.8012*****
Obstetrical factors
Induction of labour1.281.00–1.640.0540
Oxytocin augmentation******1.501.14–1.960.0053
Epidural1.140.81–1.600.4679
Episiotomy0.870.63–1.210.4043
Vacuum extraction3.322.35–4.72<0.0001
Forceps13.100.82–209.800.1025
Shoulder dystocia3.742.17–6.44<0.0001
Table 3. Multivariate analysis: risk factors of recurrent anal1 sphincter rupture (ASR)
 aOR95% CI P
  1. Mutually adjusted odds ratios (aORs) presented with 95% confidence intervals (95% CIs), based on n = 7221 patients with a complete data for this analysis (515 patients with recurrent ASR).

  2. *Overall P value = 0.0488.

Maternal factors
Delivery interval (by year)1.081.02–1.150.0122
Calendar year of second delivery1.061.03–1.09<0.0001
Grade of ASR at first delivery
 Third degree1.0  
 Fourth degree1.721.28–2.290.0005
Fetal factors
Birthweight (kg)2.942.31–3.75<0.0001
Head circumference (cm)0.910.85–0.980.0142
Presentation
 Occiput anterior1.0  
 Occiput posterior1.731.14–2.630.0154*
 Other presentations0.910.48–1.710.7639*
Obstetrical factors
Vacuum extraction2.962.03–4.31<0.0001
Shoulder dystocia1.981.11–3.540.0289

The univariate logistic regression including head circumference (n = 7234) yielded an OR of 1.11 per increasing cm (95% CI 1.05–1.19, = 0.0006), indicating an increased risk of ASR with a larger head circumference. However, in the multivariate analysis, controlling for the remaining risk factors, the head circumference demonstrated a protective effect of head circumference against ASR (aOR 0.91 per increasing cm, 95% CI 0.85–0.98, = 0.0142). To illustrate the association between the risk of recurrent ASR, head circumference and birthweight, Table 4 contains the observed proportions of ASR within four categories of weight and three categories of head circumference. For each category of weight, an almost steady decreasing pattern of the proportions is seen. For each category of head circumference, a clear increasing effect of the weight category is demonstrated (i.e. for fixed birthweight, a larger head circumference is associated with a lower risk of recurrent ASR).

Table 4. Observed proportions of recurrent anal sphincter rupture (ASR) within four categories of birthweight and three categories of fetal head circumference
Head circumference/birthweight≤34 cm35–36 cm≥37 cmTotal
  1. Within each cell, the numerator is the number of patients with recurrent ASR and the denominator is the total number of patients.

≤3.5 kg50/1104 = 0.0536/1031 = 0.036/73 = 0.0892/2208 = 0.04
3.5–4.0 kg48/586 = 0.08127/1882 = 0.0729/549 = 0.05204/3017 = 0.07
4.0–4.5 kg13/83 = 0.1679/810 = 0.1068/737 = 0.09160/1630 = 0.10
≥4.5 kg1/10 = 0.1019/98 = 0.1940/268 = 0.1560/376 = 0.16
Total112/1783 = 0.06261/3821 = 0.07143/1627 = 0.09516/7231 = 0.07

In patients with recurrent ASR, 48% (246 out of 516, 95% CI 43–52%) had one or more risk factors (vacuum extraction, excessive birthweight > 4.0 kg, shoulder dystocia or small head circumference < 32 cm). If one or more of the aforementioned risk factors were present, the risk of recurrent ASR was 11% (246 out of 2190, 95% CI 10–13%).

In the time period 1997–2010, 3129 patients had an ASR at their first delivery and had a caesarean section at their second delivery (29.9%). By comparing the group of patients who deliver vaginally with the group of patients who deliver by caesarean section after an ASR at the first delivery, we find a difference between the two groups regarding maternal age, delivery interval, calendar year of second delivery, grade of ASR in first delivery, birthweight, head circumference and gestational age (data not shown).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

The risk of recurrent ASR (OR 5.9, 95% CI 5.3–6.5) was similar to that recorded in previous studies (OR 2.5–7.7),5–11 even though we found a higher incidence rate (7.1%, 95% CI 6.5–7.7%) than was recorded in Norway (5.6%11) and Sweden (4.4%5). We have information regarding ASR in the period 1997–2010, whereas the other Nordic studies investigated recurrent ASR in the period 1967–200411 and 1973–19975. We know that the occurrence rate of ASR is increasing in the Nordic countries;19 it is possible that our analyses from a more recent time period reflect this trend, and consequently give a higher incidence of recurrent ASR.

In the univariate logistic regression model, the head circumference was a risk factor of recurrent ASR (OR 1.11 per increasing cm, 95% CI 1.05–1.19, = 0.0006), whereas in the multivariate analysis the head circumference was a protective factor against recurrent ASR (aOR 0.91 per increasing cm, 95% CI 0.85–0.98, = 0.0142). There is a positive correlation between head circumference and birthweight (Spearman’s rank correlation 0.594, 95% CI 0.58–0.61, < 0.0001), such that a large birthweight is associated with a large head circumference. However, adjusting for both risk factors in the multivariate analyses, both factors remained significant, and the birthweight confounded the relationship between head circumference and risk of recurrent ASR, a finding that was also illustrated in Table 4 containing the raw proportions of recurrent ASR in various groups of birthweight and head circumference.

The protective effect of a larger head circumference on the risk of recurrent ASR could possibly be explained by the speed of crowning. A larger head circumference is associated with a slow speed of crowning because of the mechanical conditions. In a second vaginal delivery, the expulsive phase is faster and therefore it is plausible that a large head circumference is protective. It is known that factors that increase the speed of crowning (such as vacuum extraction or forceps) imply a higher risk of recurrent ASR.6,8,12 In clinical practice we also find that a large proportion of ASRs occur after the delivery of the head, when the posterior shoulder is delivered. A smaller head circumference results in a relatively large body/shoulders (similar to macrosomia), which could increase the risk of recurrent ASR. It is also possible that a smaller head circumference fails to dilate the pelvic muscles adequately for the body following after. We found no interaction between head circumference and shoulder dystocia.

Delayed second delivery was a risk factor of recurrent ASR, with the risk of recurrent ASR increasing with the number of years between the two deliveries. It is possible that patients with problems such as flatus, stool incontinence or a traumatic delivery experience tend to postpone the second delivery. In the Danish guidelines patients with any symptoms of anal incontinence should be offered an elective caesarean section,20 but it is known that patients tend to withhold information about problems with anal incontinence.4 At the same time, the calendar year of second delivery is also associated with a higher risk of recurrent ASR, which is in accordance with a previous study describing a significant increasing incidence of ASR in the Nordic countries.19

In this study, birthweight (aOR 2.94 per increasing kg, 95% CI 2.31–3.75, < 0.0001), vacuum extraction (aOR 2.96, 95% CI 2.03–4.31, < 0.0001) and shoulder dystocia (aOR 1.98, 95% CI 1.113.54, P = 0.0289) are important risk factors for recurrent ASR, which is in accordance with previous findings [excessive birthweight (various definitions), OR 1.7–16.9;5–7,11 vacuum extraction, OR 1.4–6.5;6,8,9,11,12 and shoulder dystocia, OR 2.6–11.17]. Previous studies find that the use of forceps (OR 2.5–5.16,8,11,12) and midline episiotomy (OR 8.5–17.47,8) are significant risk factors. Forceps are rarely used in Denmark (only for two patients in our study population), and therefore we cannot estimate the effect of forceps. All episiotomies are mediolateral in Denmark, and we found no association between episiotomies and recurrent ASR.

A fourth-degree ASR at first delivery is associated with an increased risk of recurrent ASR (aOR 1.72, 95% CI 1.28–2.29). As a larger ASR results in more scar tissue, it is natural to believe that this increases the risk of recurrent ASR.

The strength of our study is the large population-based cohort, which includes a broad variety of factors that can influence the risk of recurrent ASR. As we use data from the Danish MBR we avoid recall bias, but a few factors such as the pudendal block, given before delivery, and prolonged second stage of labour could not be assessed, as they are not registered in the Danish MBR. We do not have information about symptoms of anal incontinence in the period between the two deliveries, which is a limitation of our study. Anal incontinence and other physical or psychological effects from the first ASR may influence the time from the first to the second delivery, as well as the preferred mode of delivery. This information should be considered when using risk factors to decide on preferred mode of delivery.

In this study we did not include women with caesarean section after a first delivery with ASR, and therefore it is possible that the results regarding risk factors could be biased. The patients are selected for either vaginal delivery or caesarean section (elective or emergency) on the basis of factors that we have no knowledge of. We found that patients with a caesarean section after a first delivery with ASR differ from patients with vaginal delivery in some points (maternal age, birthweight, grade of ASR in first delivery etc., data not shown). Therefore, our findings should be interpreted with caution when counselling patients with prior ASR regarding subsequent mode of delivery. It is also important to remember that more than half of the patients with recurrent ASR have none of the risk factors we have identified.

Vacuum extraction is the only modifiable risk factor we have identified, and therefore we suggest that this should be avoided in patients with vaginal delivery after prior ASR, especially when expecting a child with birthweight above 4.5 kg. However, no studies on short- and long-term consequences of late caesarean section (increased risk of bleeding, problems with delivery, low Apgar score, etc.), compared with vaginal delivery, with or without recurrent ASR, have yet been performed. Patients with prior ASR should be carefully informed when attempting vaginal delivery, in order to be aware of the risk factors associated with an increased risk of recurrent ASR.

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

The incidence of recurrent ASR in our study population was 7.1%. In addition to the well-known influence of birthweight (per increasing kg), vacuum extraction and shoulder dystocia, we found that longer delivery interval, increasing calendar year of second delivery and fourth-degree ASR at first delivery increased the risk of recurrent ASR, whereas head circumference (per increasing cm) was protective. Half of the patients with a recurrent ASR have risk factors, but most factors are first known at the very end of delivery. We need studies to quantify the adverse effects of vaginal delivery after ASR in order to improve counselling and decision-making on the subsequent mode of delivery.

Contribution to authorship

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

H.J. analysed the data and wrote the article. J.L.R. and A.S. contributed to the data analysis, interpretation of the results and writing of the article. S.R. performed the statistical analyses.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Disclosure of interests
  9. Contribution to authorship
  10. Details of ethics approval
  11. Funding
  12. Acknowledgements
  13. References
  14. Commentary on ‘Risk factors of recurrent anal sphincter ruptures: a population-based cohort study’

Anal sphincter ruptures (ASRs) are an important source of possibly preventable anal incontinence following vaginal birth.1 It is therefore clinically important to identify those women who are at greatest risk of both ASR and recurrent ASR. A poor prognosis might, for instance, be used to identify women who should avoid subsequent vaginal delivery and should instead deliver by caesarean section. The development of a valid prognostic model is dependent on many factors, including robust, prospectively collected, large data sets, with minimum missing data and minimal likelihood of measurement error or reporting biases, and focusing attention on large effects which are clinically plausible.

One of the main challenges in prognostic modelling is the identification of important prognostic variables from a large recorded number of patient characteristics, and once identified, providing unbiased estimates of prognostic value, uncertainty and model performance. When performing multiple tests, which includes testing the predictive value of many prognostic variables, the likelihood of selecting characteristics that are not important as prognostic factors (i.e. in error) quickly increases above the nominal value of 5%. Furthermore, estimates of how well the model performs are often too optimistic as a result of over-fitting to the sample at hand, but with poor generalisablity to external populations.

A number of measures can be taken to mitigate over-fitting, optimism and the likelihood of selecting a variable as being prognostic when in fact it is not. One approach is to limit the number of patient characteristics to an a priori list of clinically important variables, as opposed to selecting variables using stepwise selection procedures. Internal and external validation techniques can be used to lessen the effects of over-fitting and optimism penetrating into clinical practice.2 External validation evaluates how the model performs on other independent samples. Unfortunately, many prognostic models are developed without recourse to validation on independent samples.3 Internal validation is a complementary method, and although not a substitute for external validation, requires the use of only the primary data. One well-known but now outdated internal validation technique is to split the sample into two, using one sample for model fitting and the other for validation. More robust methods, known as shrinkage, use bootstrap samples to repeatedly sample with replacement from the current data set, re-fitting the prognostic model each time and averaging results from fitted prognostic models over the repeated samples.2 These methods result in a shrinkage factor for the regression coefficients, pulling estimated effects towards the null value (i.e. value of no effect).

In the accompanying paper, Jango and colleagues have used a large (n = 7336) prospectively collected database to identify prognostic variables for recurrent ASR. They started with an a priori list of 16 clinically important factors and, through forward and backward selection procedures, identified which of these were statistically significant in a multivariate model. Many of the identified characteristics, such as birthweight, were already known to be of clinical importance. However, one characteristic, head circumference, was estimated to be a protective factor in the fully adjusted model (OR 0.91 per increasing cm, 95% CI 0.85–0.98), which is contrary to previously held beliefs. The authors, admirably, considered the sensitivity of results to shrinkage, estimating the shrinkage factor to be 0.99, which suggests that over-fitting is probably not of concern here. Whereas the authors provide clinical plausibility for the protective effect of increased head circumference (increased crowning time), the estimated effect size is relatively small (OR = 0.91), and a moderately large number of patient characteristics were candidates in the prediction model. Before large head circumference becomes widely known in clinical communities as being a protective factor against ASR, external validation should be used to validate this finding, mitigating the likelihood of obstetricians and gynaecologists becoming influenced by the increasing numbers of false-positive research findings.4

Finally, it is perhaps worth noting that although software packages like stata (StataCorp, College Station, Texas, USA) and sas (Sas Institute Inc, Cary NC, USA) have facilitated the implementation of these methods, statistical expertise should be sought, as with increasingly complex methods the potential for misuse also increases.

K Hemming

Public Health, Epidemiology and Biostatistics,

University of Birmingham, Birmingham, B15 2TT, UK

References

1 Jangö H, Langhoff-Roos J, Rosthøj S, Sakseb A. Risk factors of recurrent anal sphincter ruptures: a population-based cohort study. BJOG 2012; DOI:10.1111/j.1471-0528.2012.0486.x.

2 Steyerberg EW. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updatin. New York, USA: Springer, 2008.

3 Bouwmeester W, Zuithoff NP, Mallett S, Geerlings MI, Vergouwe Y, Steyerberg EW, et al. Reporting and methods in clinical prediction research: a systematic review. PLoS Med. 2012;9:e1001221.

4 Ioannidis JP. Why most published research findings are false. PLoS Med 2005;2:e124.