**4.2 Information gathering**

Information on UI in pregnancy and postpartum is often gathered through questionnaires, but objective testing (Morkved & Bo, 1999), personal structured interviews (Chiarelli & Campbell, 1997, Morkved & Bo, 1999) or semi structured interviews (Farrell et al., 2001, Spellacy, 2001) or phone interviews (Baydock et al., 2009) by doctors or assistants, or reviews of existing medical records (Spellacy, 2001) are also used. Information collected by interview makes it possible to clarify and gather more and better information regarding UI. Thom found higher prevalence figures of UI when data was gathered by structured interview compared to questionnaire (Chiarelli & Campbell, 1997, Thom, 1998). Medical records often lack important information, leading to low prevalence estimates. Studies have found low agreement between self reported UI and clinical assessment (Diokno et al., 1988, Milsom et al., 1993). A review on variations in estimates of UI found that objective testing according to the "UI sign" definition led to lower prevalence estimates than questionnaire based studies using the "UI symptom" definition (Thom, 1998).

#### **4.3 Type of study**

A large proportion of studies on UI in pregnancy or postpartum are cross sectional (Table 1 – 4) or retrospective. If a woman has UI when answering a retrospective study, this may affect her reporting of UI by improving her memory about earlier UI leading to a recall bias. Cross sectional studies have less valid incidence figures than prospective cohorts. Crosssectional studies can gather information about the prevalence of UI, but they cannot distinguish between incident and long-established UI. Therefore, cross-sectional studies can usually only measure prevalence of UI. Also, they cannot identify cause-and-effect relationships as exposure and outcome information are gathered at the same time.

### **4.4 Timing of data collection**

30 Urinary Incontinence

A wide range of prevalence estimates of UI in pregnancy and postpartum have been presented. There are several methodological reasons for these diverging incidence and





The ICS definitions and terminologies of UI according to the above descriptions have been revised several times (Abrams et al., 1988, Abrams et al., 2002, Haylen et al., 2010). The current definition of UI symptoms is "Complaint of involuntary loss of urine" (Haylen et al., 2010). In the 2002 definition, UI symptoms were not enough to set the UI diagnose; UI signs were needed. Today the majority of studies on UI define UI according to UI symptoms. Studies on UI have used the definitions at the time. As definitions change, prevalence

Information on UI in pregnancy and postpartum is often gathered through questionnaires, but objective testing (Morkved & Bo, 1999), personal structured interviews (Chiarelli & Campbell, 1997, Morkved & Bo, 1999) or semi structured interviews (Farrell et al., 2001, Spellacy, 2001) or phone interviews (Baydock et al., 2009) by doctors or assistants, or reviews of existing medical records (Spellacy, 2001) are also used. Information collected by interview makes it possible to clarify and gather more and better information regarding UI. Thom found higher prevalence figures of UI when data was gathered by structured interview compared to questionnaire (Chiarelli & Campbell, 1997, Thom, 1998). Medical records often lack important information, leading to low prevalence estimates. Studies have found low agreement between self reported UI and clinical assessment (Diokno et al., 1988, Milsom et al., 1993). A review on variations in estimates of UI found that objective testing according to the "UI sign" definition led to lower prevalence estimates than questionnaire

A large proportion of studies on UI in pregnancy or postpartum are cross sectional (Table 1 – 4) or retrospective. If a woman has UI when answering a retrospective study, this may

(Abrams et al., 1988, Abrams et al., 2002, Haylen et al., 2010)

based studies using the "UI symptom" definition (Thom, 1998).

1988, Abrams et al., 2002, Haylen et al., 2010)

**4. Why do estimates differ?** 

The concept of UI can be based on:

processes) (Abrams et al., 2002)

estimates will also change.

**4.2 Information gathering** 

**4.3 Type of study** 

prevalence estimates.

**4.1 UI definition** 

2002)

Timing of data collection can affect prevalence estimates of UI in pregnancy. Some studies question women about UI during each trimester, but most studies question women at one certain time point in pregnancy (Brown et al., 2010, Lewicky-Gaupp et al., 2008) or just after birth (Sottner et al., 2006). Some studies do not report what time in pregnancy the women reported UI (Sharma et al., 2009). As prevalence of UI increases in pregnancy, the time of information gathering will affect the prevalence estimates of UI during pregnancy. When it comes to data collection postpartum, some studies report on UI at 6 - 9 weeks postpartum (D'Alfonso et al., 2006, Lewicky-Gaupp et al., 2008, Meyer et al., 1998), 3 months (Eason et al., 2004, Hannah et al., 2002), 4 months (Baydock et al., 2009), 6 months (Thomason et al., 2007), 12 months (Serati et al., 2008) or > 12 months (Foldspang et al., 2004, Fritel et al., 2004) postpartum. The time of information gathering postpartum might affect incidence and prevalence estimates of UI. However, a recent review indicates that prevalence of UI is stable first year postpartum (Thom & Rortveit, 2010), and time of data collection postpartum may therefore be of less importance.

#### **4.5 Threshold**

Permanence, frequency and volume are used by authors as threshold to define women with UI in association with pregnancy. Permanence or duration can be defined as one or more episodes of UI in the previous month (Brown et al., 2010, Wilson et al., 1996). Some authors use longer periods, like trimesters (Schytt et al., 2004) or the 6 months postpartum period (Schytt et al., 2004). Some authors investigate severe UI defined by weekly or daily leakage (Al-Mehaisen et al., 2009) while others do not report any cut-off (van Brummen et al., 2006). Prevalence estimates are lower for daily UI compared to weekly or monthly UI (Thom, 1998). Some studies have a cut-off for minimum frequency, amount or severity of UI for women to be included in the study as incontinent. A high cut-off decreases the number of women who fulfil the UI criteria in a study. Differing thresholds may explain differing incidence and prevalence estimates of UI.

#### **4.6 Type of UI**

Stress UI predominates in young women. Stress UI is more common in pregnancy and postpartum than urgency UI and mixed UI. Also, the incidence of pure urgency UI in pregnancy or postpartum is low compared with incidence of stress UI and mixed UI. The prevalence of pure stress UI is reported to be 2 – 8 times higher than the prevalence of pure urgency UI in pregnancy (Brown et al., 2010, Goldberg et al., 2005, Raza-Khan et al., 2006). Prevalence of mixed UI is reported to be 0.3 – 1.5 times of the prevalence of pure stress UI in pregnancy (Brown et al., 2010, Goldberg et al., 2005, Raza-Khan et al., 2006). The

Epidemiology of Urinary Incontinence in Pregnancy and Postpartum 33

It is essential to research on UI that incidence and prevalence estimates can be properly assessed and recorded. As clinicians objective testing and patients' symptoms often differ in their perspective of UI (Milsom et al, 2009), the use of questionnaires to approach patients symptoms are more used recent years. There are an increasing number of questionnaires to assess UI. The Symptom and Quality of Life Committee of the International Consultation on Incontinence performed a systematic review of questionnaires related to urinary incontinence (Avery et al, 2007). They identified 17 questionnaires on UI in women (assessing symptoms, quality of life or both) that were highly recommended; that is questionnaires that were seen as an established measure with documented, rigorous validity, reliability and responsiveness in several clinical studies. However, only 38 % of all clinical trials use these questionnaires (Avery et al, 2007). The proportion in descriptive studies using robust validated questionnaires is likely to be even lower. Some of the variability in UI incidence and prevalence estimates is likely to be related to the range of

All the above methodological factors can influence UI estimates in a study. Unfortunately we do not know all factors that influence UI estimates. Some variation in prevalence

Reported incidence and prevalence estimates in pregnancy and postpartum vary (Table 5). Incidence of UI is high during pregnancy. Close to 50 % of all women experience UI during pregnancy. Delivery is one of many factors that lead to a high incidence of UI postpartum. About 1/3 women experience UI postpartum. This is the first review trying to summarize

**Time point Source of data Primiparous Parous** 

11 – 50 % 17 – 45 %

4 – 70 % 35 – 55 %

0 – 26 % 5 – 21 %

6 – 67 % 15 – 45 %

29 %

8 – 57 % 45 – 54 %

14 – 85 % 24 – 67 % 32 – 64 %

4 – 21 % 8 – 15 % 7 – 15 %

3 – 45 % 18 – 38 % 15 – 30 % 9 – 31 % 33 %

Narrow range Table 1

Narrow range Table 2 Report (Milsom et al., 2009)

Narrow range Table 3 Review (Nygaard, 2006)

Narrow range Table 4 Report (Milsom et al., 2009) Review (Nygaard, 2006) Systematic review (Thom &

Rortveit, 2010)

Table 5. Range of incidence and prevalence estimates for any UI by parity.

**4.9 Questionnaire** 

different questionnaires used.

**5. Conclusion** 

estimates between studies will always remain.

the UI estimates in association with pregnancy.

Incidence in pregnancy Range Table 1

Prevalence in pregnancy Range Table 2

Incidence postpartum Range Table 3

Prevalence postpartum Range Table 4

stress/urgency ratio is reduced postpartum as prevalence of stress UI decline. Several studies focus solely on stress UI (Mason et al., 1999, Torrisi et al., 2007, Viktrup et al., 1992). Prevalence figures in these studies are likely to be lower than in studies that include both urgency UI and mixed UI in their analyses (Thom, 1998).

#### **4.7 Characteristics of study population**

The study population influences prevalence of UI. Some studies on UI in association with pregnancy use study populations from tertiary care hospitals (Baydock et al., 2009, Raza-Khan et al., 2006), leading to recruitment of highly selected participants. BMI distribution, age distribution, parity distribution, proportion of European or Hispanic population, proportion of women having vaginal delivery all influence prevalence figures of UI. Mothers BMI and age at first delivery have risen the 50 years. These demographic variables might partly explain why studies from 1970-1980 tend to report lower UI estimates compared to recent studies. Some studies include only women having SVD (Altman et al., 2006, Arrue et al., 2010, Baydock et al., 2009), which will give a higher prevalence estimate of UI than if the study also had included CS. Many studies on UI in association with pregnancy either adjust or report stratified analyses for age (Solans-Domenech et al., 2010), BMI (Eason et al., 2004), race (Connolly et al., 2007) and mode of delivery (Eason et al., 2004). Effect estimates are thereby controlled for baseline imbalances in these important patient characteristics. However, dissimilar use of statistical stratification and adjustment makes it difficult to compare findings. Pooled prevalences figures can be misleading and readers should be careful in generalising the findings to a population outside the study population.

#### **4.8 Bias**

Many studies on UI in pregnancy try to gather information from all pregnant women in the community; so called population based studies (Boyles et al., 2009, Thompson et al., 2002). Participation rates for epidemiologic studies have been declining during the decades with even steeper declines in recent years (Galea and Tracy, 2007). Several large surveillance surveys in USA report overall decrease in participation rates well below 50 %. Population based studies on UI in association with pregnancy are challenged with the same problems. Boyles reported a response rate of 39 % (Boyles et al., 2009), Wesnes of 45 % (Wesnes et al., 2007). These studies are prone to a biased response rates/selection bias which may invalidate the prevalence estimates. Nulliparous women are more likely to participate and tell their pregnancy stories in studies compared with parous women (Magnus et al., 2006). Declining participation rates and the growing complexity of reasons for study nonparticipation add unpredictability about who is participating in a study and who is not. It challenges the ability of these studies to confidently obtain a population-representative sample (Galea and Tracy, 2007).

Known differences between responders and non-responders may be compensated during analyses. The major problem is unknown response bias, such as the possibility of different response rates between continent and incontinent women (Cartwright, 1983). Due to embarrassment and feeling uncomfortable about reporting UI, incontinent women may deny or not answer questions about UI. Conversely, incontinent women may find the subject particularly relevant, and therefore respond to a greater extent than continent women. At present, we do not know how these factors may affect the response rates. To minimize selection bias one should always aim at the highest possible response rates.

#### **4.9 Questionnaire**

32 Urinary Incontinence

stress/urgency ratio is reduced postpartum as prevalence of stress UI decline. Several studies focus solely on stress UI (Mason et al., 1999, Torrisi et al., 2007, Viktrup et al., 1992). Prevalence figures in these studies are likely to be lower than in studies that include both

The study population influences prevalence of UI. Some studies on UI in association with pregnancy use study populations from tertiary care hospitals (Baydock et al., 2009, Raza-Khan et al., 2006), leading to recruitment of highly selected participants. BMI distribution, age distribution, parity distribution, proportion of European or Hispanic population, proportion of women having vaginal delivery all influence prevalence figures of UI. Mothers BMI and age at first delivery have risen the 50 years. These demographic variables might partly explain why studies from 1970-1980 tend to report lower UI estimates compared to recent studies. Some studies include only women having SVD (Altman et al., 2006, Arrue et al., 2010, Baydock et al., 2009), which will give a higher prevalence estimate of UI than if the study also had included CS. Many studies on UI in association with pregnancy either adjust or report stratified analyses for age (Solans-Domenech et al., 2010), BMI (Eason et al., 2004), race (Connolly et al., 2007) and mode of delivery (Eason et al., 2004). Effect estimates are thereby controlled for baseline imbalances in these important patient characteristics. However, dissimilar use of statistical stratification and adjustment makes it difficult to compare findings. Pooled prevalences figures can be misleading and readers should be careful in generalising the findings to a population outside the study population.

Many studies on UI in pregnancy try to gather information from all pregnant women in the community; so called population based studies (Boyles et al., 2009, Thompson et al., 2002). Participation rates for epidemiologic studies have been declining during the decades with even steeper declines in recent years (Galea and Tracy, 2007). Several large surveillance surveys in USA report overall decrease in participation rates well below 50 %. Population based studies on UI in association with pregnancy are challenged with the same problems. Boyles reported a response rate of 39 % (Boyles et al., 2009), Wesnes of 45 % (Wesnes et al., 2007). These studies are prone to a biased response rates/selection bias which may invalidate the prevalence estimates. Nulliparous women are more likely to participate and tell their pregnancy stories in studies compared with parous women (Magnus et al., 2006). Declining participation rates and the growing complexity of reasons for study nonparticipation add unpredictability about who is participating in a study and who is not. It challenges the ability of these studies to confidently obtain a population-representative

Known differences between responders and non-responders may be compensated during analyses. The major problem is unknown response bias, such as the possibility of different response rates between continent and incontinent women (Cartwright, 1983). Due to embarrassment and feeling uncomfortable about reporting UI, incontinent women may deny or not answer questions about UI. Conversely, incontinent women may find the subject particularly relevant, and therefore respond to a greater extent than continent women. At present, we do not know how these factors may affect the response rates. To minimize selection bias one should always aim at the highest possible response rates.

urgency UI and mixed UI in their analyses (Thom, 1998).

**4.7 Characteristics of study population** 

**4.8 Bias** 

sample (Galea and Tracy, 2007).

It is essential to research on UI that incidence and prevalence estimates can be properly assessed and recorded. As clinicians objective testing and patients' symptoms often differ in their perspective of UI (Milsom et al, 2009), the use of questionnaires to approach patients symptoms are more used recent years. There are an increasing number of questionnaires to assess UI. The Symptom and Quality of Life Committee of the International Consultation on Incontinence performed a systematic review of questionnaires related to urinary incontinence (Avery et al, 2007). They identified 17 questionnaires on UI in women (assessing symptoms, quality of life or both) that were highly recommended; that is questionnaires that were seen as an established measure with documented, rigorous validity, reliability and responsiveness in several clinical studies. However, only 38 % of all clinical trials use these questionnaires (Avery et al, 2007). The proportion in descriptive studies using robust validated questionnaires is likely to be even lower. Some of the variability in UI incidence and prevalence estimates is likely to be related to the range of different questionnaires used.

All the above methodological factors can influence UI estimates in a study. Unfortunately we do not know all factors that influence UI estimates. Some variation in prevalence estimates between studies will always remain.
