Evaluation of the psychometric properties of the London Measure of Unplanned Pregnancy in Brazilian Portuguese

For translation purposes, the English LMUP was sent to two native Brazilian Portuguese speakers (both professors at the University of Sao Paulo, with research expertise in reproductive health, and aware of the purpose and background of the LMUP) who each independently translated it into Portuguese. ALVB reviewed the translations and discussed the differences at a consensus meeting with four other health researchers – doctors and nurses (two experts in psychometrics and two in reproductive health). At the time of translation, no content of the UK version was changed. The agreed translation produced by this meeting was sent for back-translation to a native British English speaker who spoke Portuguese fluently as a second language. This person was only broadly aware of the purpose of the LMUP.

Following back-translation, we piloted the Portuguese LMUP in the form of a self-completion survey with 126 pregnant women waiting for their first prenatal care consultation in a primary health care service in a medium sized municipality in Sao Paulo state, Brazil. We asked the women, “Have you understood the questions of this questionnaire?”, with the answer options “I did not understand anything”; I understood it a bit”; “I understood almost everything, but still have some doubts”; and “I understood it perfectly well and I have no doubts”. Only one woman expressed doubts about the questions but made no suggestions for improvement. Although the majority reported they could understand it all, some women who had had low scores on items 1-5 (and therefore appeared to have an unplanned pregnancy) also reported that they had carried out a pre-pregnancy preparatory behavior (a higher score on item 6). In our initial reliability analyses we could also see that item 6 had an item-rest correlation of <0.2 and was performing differently to the other items within the scale. We were concerned that item was being miscomprehended by some women. Generally, Brazilian women seek health assistance before conception just when they notice infertility problems or when they have pre-existing diseases, such as diabetes or hypertension. For instance, pre-pregnancy folic acid use has previously been reported by only 4.3 % of Southern Brazilian women, which denotes the scarce amount of preconception care delivered in Brazil [26]. We also considered the heterogeneous schooling profile of Brazilian health system users, with many poorly educated women. Considering this context, we decided to not change the response options because we considered them clear and appropriate for Brazil. Instead, we adapted the Brazilian Portuguese LMUP for interviewer-administration, in the same way the Indian and Malawian versions did [18, 19], with changes in the wording in a way that made its administration by an interviewer possible. ALVB and FBC adjusted the phrasing and NCN and OAS checked the new interviewer-administered version, which satisfactorily addressed this issue.

The second setting (setting 2) comprised 12 public primary health care services. Every pregnant woman, also irrespective of pregnancy trimester, waiting for prenatal care, vaccination, or educational group meetings on specific weekdays from April to July 2013 was invited to participate in the study (n = 513). The purpose was to understand the reasons women with unplanned pregnancy had not used emergency contraception to prevent the pregnancy [27].

The third setting (setting 3) was another public maternity unit. Women hospitalized due to incomplete abortion – spontaneous or induced – were invited to participate in the study after emergency treatment, while waiting for hospital discharge, from May to August 2012 (n = 170). Having a setting with women who had an incomplete abortion provided us with two conditions essential to testing LMUP scale: first, a pregnancy outcome that is not a birth; second, a group of women who would include some with low pregnancy intention, as this group consisted of women with spontaneous and induced abortion. We did not ask women if the abortion was induced as we recognized it would be inaccurately reported due to the restrictive abortion laws and therefore underestimated. Women who responded to the LMUP at this setting were then followed-up for six months to understand their contraceptive behavior and its relation to primary health care [23].

Although we had no refusals in settings 1 and 2, we had 14 refusals in setting 3. Considering the restrictive laws on abortion in Brazil, we believe some women might have felt apprehensive taking part in this study while they were hospitalized for post-abortion care.

Nurses and midwives, who were graduate students, conducted all the interviews after training. Women were interviewed in quiet places in the clinics. In each setting, all pregnant/postpartum women/women who had an abortion attending the services were invited to participate in the study. In addition to the LMUP, women answered structured questions about their demographic and reproductive characteristics.

We entered data twice on Epiinfo software and corrected typing mistakes. The analysis of psychometric properties were conducted with Stata 13.0 using a Classical Test Theory-based approach to facilitate comparison with the original UK study [10] and previous validations [16‐20].

Rates of missing data were assessed to give an indication of items that might have a problem with acceptability or comprehension. To assess item discrimination the item-endorsement values were examined to see if any item response option had an endorsement greater than 80 %. The distribution of total scores was considered to see if the full range of scores were present and to evaluate the targeting of the scale.

To assess reliability, internal consistency was evaluated by calculating the Cronbach’s α statistic (>0.7 indicating acceptable reliability) [28] and the item-rest correlations (<0.2 indicating that the item is contributing little to the homogeneity of the scale) [21]. We also examined the inter-item correlations, checking that all correlations were positive.

Consistent with the UK LMUP development and evaluation study [10], construct validity was examined using two methods: principal component analysis and hypothesis testing. We used principal component analysis (without rotation, requesting as many components as there were Eigenvalues >1) to test the hypothesis that all items would load onto one component (i.e. measure the same construct). Hypotheses related to pregnancy intention were generated from information provided by the Brazil DHS 2006. We tested four hypotheses: 1) women in the extreme age groups (younger and older) would have lower levels of pregnancy intention/lower LMUP scores; 2) women from lower educational backgrounds would have lower pregnancy intention/lower LMUP scores; 3) those who are not married would have lower intention/lower LMUP scores; and 4) women from the third setting (i.e. those who were hospitalized for incomplete abortion and therefore likely to include some with induced abortions) would have lower intention/lower LMUP scores compared with the other two centres. We used the Wilcoxon Rank-Sum test and the Mann Whitney U test to assess significance.

Finally, in keeping with the USA evaluation of the LMUP [17], we carried out an exploratory analysis based on the principles of modern test theory: a Mokken scaling procedure (monotone homogeneity assumption). Mokken models are non-parametric models within the family of Item Response Theory models. The results of the Mokken analysis allowed us to see whether the items conformed to a probalistic Guttman structure, i.e. that items vary in ‘difficulty’, some being easy to endorse, some being more difficult to endorse, and that respondents who have a particular level of the construct (in this case pregnancy planning/intention) should broadly endorse items up to the level of their construct and then not endorse items beyond that. The Loevinger H coefficient produced by the Mokken analysis relates to Guttman errors, with a lower H value indicating more observed Guttman errors. Items with a Loevinger H coefficient >0.3 were eligible for scaling [29, 30]. The scale as a whole was also assessed by a Loevinger H coefficient, with <0.4 meaning the scale is “weak”, 0.4 to 0.49 meaning the scale is “medium”, and >0.5 meaning the scale is “strong” [29]. The construction of an adequate Mokken scale confirms that the raw score can be used to order respondents on the construct being measured [30].