The search strategy was informed by the earlier scoping review that examined the extent, range, and nature of the evidence on measuring ITU ⁵ . This scoping review indicated that further analysis was needed to better understand whether ITU has significant effects on subsequent contraceptive uptake, so we performed a systematic review to examine this relationship. For this systematic review, we followed the PRISMA guidelines for reporting systematic reviews and meta-analyses ²⁶ . Please see Figure 1. We searched PubMed, PsycInfo, Web of Science, and the Cochrane Collaboration for studies published between 1975 and August 2020 using search terms relevant to intent-to-use and contraceptive use. The search terms and strategy are shown in the protocol ²⁵ .

The study design included in the review were experimental, quasi-experimental, or observational studies with either a pre/post or treatment/control comparison. Studies were eligible for inclusion if they: (1) examined contraceptive behaviour (excluding condom use only), (2) included disaggregated integral measures of ITU contraceptives and later contraceptive use, (3) included at least one quantitative measure of the association between ITU contraceptives and actual contraceptive use, (4) the study population was women of reproductive age, (5) were peer-reviewed, and (6) were written in the English language. There were no limits to study inclusion based on the study setting. Studies were excluded if the full text was not accessible, not published in a journal (e.g., dissertations), or not written in English.

We exported the search results into Endnote21 to remove duplicates and then imported the de-duplicated results into Excel 2021. Two authors (VB and SE) independently screened 1,464 titles and abstracts ²⁷ . Where discrepancies arose, the authors resolved disagreements through discussion between the reviewers. Subsequently, SE and VB independently reviewed 39 full-text articles to ascertain their eligibility for inclusion and resolved disagreements through discussion. Data extracted included the year of publication, study purpose, location, study design, sample size, participant characteristics, follow-up period in months, type of contraceptive used, measurement of ITU, measurement of contraceptive use, attrition, number of participants who reported ITU contraception who subsequently did and did not use contraception, the number of participants who reported no ITU contraception who then did and did not use contraception, and effect measure and size (See Table 1). Data were then independently extracted from the 10 included articles by one author (SE) using a predesigned data extraction form ²⁷ . One author (KW) reviewed the full papers and checked the data extraction. We calculated unadjusted odds ratios for the included studies, as several did not report adjusted odds ratios for the relationship between ITU and contraceptive use. We report both our calculations of the unadjusted odds ratios and author’s adjusted odds ratios with the variables adjusted for in our presented results.

One author (SE) assessed the risk of bias using the Joanna Briggs Institute Critical Appraisal Checklist for Cohort Studies ²⁸ , which assesses the trustworthiness, relevance and results of cohort studies. A scoring system assigns a score of 1 or 0 against each risk of bias domain. The scores were assigned and then summed across each domain, and studies were given a score ranging from 1 to 11. Subsequently, studies were classified into low (score below 5), medium (score of 6 to 8) and high quality (score above 8). Table 2 outlines the results of the assessment for each study.

Although some of the included papers did report relationships between intention to use and contraceptive use adjusted for a variety of covariates, these covariates are not the same across different studies. This means that either different studies included completely different covariates in their adjusted models or the way similar covariates were measured was not comparable across studies. Therefore, we calculated unadjusted odds ratios for the relationship between ITU and contraceptive use and reported on the adjusted ratios reported by authors. Despite the small sample size, we attempted to run a meta-analysis that combined the results of the studies for which we were able to calculate unadjusted odds ratios, as this would have generated a more robust source of evidence. However, meta-analysis diagnostics indicated that the high degree of variation across studies in follow up times, predictor and outcome measures, and sample populations (See Table 2) precluded pooling the data for a meta-analysis. This is the first attempt to systematically synthesise this information, and more studies that assess the longer-term relationship between reported intent to use and contraceptive use are needed for any future meta-analyses.

The search yielded 1,464 articles. Many papers were excluded because they did not have a clear definition of intention to use (732), did not state an association between intention to use and contraceptive use (235), did not meet the study design requirements (238), did not contain sufficient information in the text to be assessed against the inclusion criteria (30), focused on condoms (161), did not include a measure of contraceptive use (61) or focused on only on the drivers of intention to use and did not test the association with actual use (17).

After the initial abstract screening and full paper review, a total of 10 articles were included ²⁷ . One of the 10 studies was conducted in the USA. The remaining studies were undertaken in low- and middle-income country (LMIC) settings: Bangladesh (n=1), Brazil (n=1), Cambodia (n=1), Ghana (n=1), India (n=1), Malawi (n=1), Morocco (n=1), Rwanda (n=1), and Uganda (n=1). All 10 studies were longitudinal cohort studies with pre-and post-tests or treatment and control groups. The characteristics of the studies, such as study aim, population, location, study design, follow up period, quality rating, effects measures, measure of ITU and measure of contraceptive use, are summarized in Table 1.

The number of participants varied between studies from 219 to 3,933, while six papers had sample sizes of approximately 200 to 300 participants. The papers looked at a variety of different participants – either women as broad category (e.g., sexually active or married) or at different points in their reproductive career (e.g., pre and post-partum). Two papers sampled married women ^{16, 17} ; two papers sampled postpartum women ^{29, 30} ; two papers sampled pregnant women ^{31, 32} and another two sampled sexually activity women ^{7, 33} . Only one paper looked at women post-abortion ³⁴ . These papers provide 28,749 person-years of data (N=10,925).

Half of the 10 included studies did not describe how exactly intention-to-use contraception was measured, and no details are provided on the exact wording of the items used to solicit information on the intention to use contraception ^{29, 31– 34} . Of the remaining studies, three used items that asked about the intention to use contraception in the future with no exact time frame specified ^{7, 16, 33} . Only one study used items that asked about intention to use contraception within a specific time; the time frame used was within the year ³⁰ .

In contrast, the majority of included studies did outline how they captured the outcome measure, contraceptive use. All of the studies used self-reported contraceptive use as the outcome measure (n=10). However, Johnson et al. used clinical records and two studies did not specify how they captured contraceptive use ^{17, 29, 32} .

There was extensive heterogeneity in the measures used to report associations or effects in the included studies. Four papers used odds ratios to examine the relation between intention-to-use and use of contraception ^{7, 16, 31, 35} . Across the studies that used odds ratios, researchers compared women who intended to use contraception to women who did not intend to use any method. These four studies found higher odds of women using contraception if they had planned to use it previously; this finding was statistically significant at p<.001 for three of the four studies. One paper used correlation coefficients ²⁹ , and two papers used hazard ratios ^{7, 30} . The remaining papers reported on their findings using “concordance” ³³ , and simple percentages or proportions ^{32, 34} .

Of the 10 studies for which we calculated unadjusted odds ratios of contraceptive use by intention to use status, six had significant, increased odds of subsequent contraceptive use after reporting an intention to do so at an earlier point, see Table 2. The unadjusted associations range from 0.75–7.40 based on odds ratios. Of the 10 included studies, five reported on an adjusted relationship between intent to use as a predictor variable and contraceptive use as an outcome variable. Of these, four found significantly increased odds or hazards of contraceptive use given stated intent to use at the initial measurement. These studies adjusted for a variety of covariates, with the most common being age, measures of the number of pregnancies, and education. As would be expected, the magnitude of significant unadjusted odds ratios generally decreases with adjustment for covariates, however the strength of the association, e.g, the precision with which the confidence intervals are estimated to be different than 1, does not. In one case, Tang et al. (2016), our unadjusted odds ratio was non-significant, while the author’s calculation of an adjusted hazard ratio was. In the study conducted by Lori et al. (2018), our unadjusted calculation was significant at the p<.05 level while the authors’ adjusted calculation is non-significant.

Two of the included papers examined only long acting reversible method (LARC) use at follow up ^{30, 32} . Three studies included only what would be considered modern contraceptive methods, including LARCS such as IUDs and implants, and shorter term methods like pills, injectables, vaginal rings, and condoms, alongside sterilization ^{29, 33, 35} . The remaining studies grouped contraceptive methods into various groupings, such as ‘modern’, ‘modern and reversible’, ‘modern and permanent’, and ‘traditional’ ^{7, 16, 17, 31, 34} .

There were also significant differences in the intervals between baseline and follow-up within the included studies. Most of the studies examined the relationship between intention to use and contraceptive use over long-term (longer than one-year) periods, ranging from one-year follow up measurements to six years in between measurements ^{7, 16, 17, 31, 34, 35} . Some of these studies of longer duration included intervening measurements at specified month-intervals ^{7, 30, 35} . The differences in odds ratios of contraceptive use at these intervals especially highlights the need for subsequent work to focus on specific intervals to better understand the duration range of intention to use reports. The remaining papers examined contraceptive use for less than one year, or the duration of follow up was unspecified ^{29, 32, 33} .

Of the 10 studies included, six focused in and around pregnancy; this refers to the antenatal, postabortion, and postpartum period. Two of the 10 studies examined intention to use contraception among women in the postpartum period and followed up on whether women’s intention had transformed into use over the following 12 months ^{4, 29, 30} . A further three studies examined women’s choice to use contraception in the antenatal period and followed up six months to one year after to see if they were using a method ^{31– 33} .

Only one study looked at the intention to use among women following an abortion ³⁵ . In Cambodia, Adelman et al., examined what characteristics collected at the point of abortion are associated with oral contraceptive use at four and 12 months after the abortion. Intention to use contraception was found to be positively associated with increased contraceptive use over the year ³⁵ .

The remaining four studies looked at the intention to use contraception among women with partners, including married women ^{7, 16, 17, 34} . Using longitudinal data from rural Bangladeshi women (n=2,500), Callahan and Becker found that intention to use a method was predictive of subsequent contraceptive use for women with and without an unmet need. Only two of these studies specified whether the women were non-users ^{7, 16, 17} . In Uganda, Sarnak et al., compared unmet need and contraceptive adoption to contraceptive intentions and use ⁷ . They found that women who intended to use contraception in the future used contraceptives significantly earlier (aHR = 1.45, 95% CI = 1.22-1.73) than those who did not intend to use contraception ⁷ . Interestingly, women with an intention to use but not classed as having no unmet need had the highest rate of adoption compared to those with no unmet need and no intention to use (aHR = 2.78, 95% CI = 1.48-5.258 ⁶ . The follow-up period to see if married women’s intentions had turned into actual contraceptive use was a one-to-three-year period in this set of studies ^{7, 16, 17, 34} .

We used the Joanna Briggs Institute Critical Appraisal Checklist for Cohort Studies ²⁸ , which assesses the trustworthiness, relevance and results of cohort studies, to rate the quality of each study using the following domains: the sample, exposure measures, confounding factors, outcome measures, follow-up time reported, and type of analysis used. Four studies were graded as high quality, and five were of medium quality. One study was classed as low quality.

Reading across the papers, there is inconsistency in how ITU is currently operationalized and applied. This analysis found that five (n=5) papers did not provide details on the wording of the items used to measure ITU ^{29, 32– 35} . Based on what information is available from the included papers, five (n=5) papers captured goal intentions ^{7, 16, 17, 24, 35} whereas four (n=4) captured implementation intention ^{20, 30, 31, 34} . This finding is significant because established behavioural theory suggests that distinguishing the type of intention may be helpful as implementation intentions are more likely to translate into the behaviour than goal intentions ³⁶ . Gollwitzer and Sheeran helpfully distinguish between goal intention and what people plan to do some time in the future ³⁷ . In contrast, implementation intentions are more specific regarding when, where, and how one's achievement of an intention will occur. Implementation intentions tend to be oriented towards a particular action, whereas goal intentions tend to be outcomes achieved by performing several actions ³⁷ . Gollwitzer and Sheeran argue that goal intentions do not prepare people for dealing with the problems they face in initiating, maintaining, disengaging from, or overextending themselves in realizing their intentions ³⁷ . In contrast, an implementation intention sets out the when, where, and how in advance and is a form of planning that bridges the intention-behaviour gap, increasing the likelihood of intentions being realized ³⁷ . Unfortunately, none of the papers included distinguished between goal and implementation intentions. Additional research on how ITU is conceptualized and operationalized is needed to understand how different types of intentions (e.g., goal vs implementation) predict contraceptive use and continuation. To address this, further research in needed using standardized ITU and outcome measures and similar follow-up durations amongst similar populations to assess the magnitude and direction of associations between ITU and contraceptive use.

Given the heterogeneity, several potential confounding variables could affect whether an intention to use contraception leads to future contraceptive use. These possible confounding variables make it difficult to establish a causal link between ITU and contraceptive use. This review points to several potential confounding variables to consider in future work.

Several studies in this review focused on populations during and around pregnancy. This could be an artefact of research study design as recruiting women attending pregnancy-related services may be easier. It could be an artefact of programme design in that women are more likely to engage in healthcare during pregnancy. Similarly, parity and relationship status may also affect whether an intention to use contraception translates into actual use. Future research should examine how pregnancy status may affect intentions to use contraception compared to women seeking to prevent pregnancy who are not pregnant.

Another variable that may affect the relationship between intention to use and actual use is the type of contraception method being considered. For example, long-acting reversible contraceptive methods may require more commitment and planning, whereas short-acting methods may be easier to access and use. Hence, the specific type of method may differentially affect the ease or difficulty of a person transforming their intentions into action. Work on developing a psychometric scale on contraceptive intent highlighted that contraceptives are a form of medication, and the woman's desire and adherence to them are influenced by beliefs about the medicine ¹⁰ . Another variable we noted is how long it may take to move from intention to action and when to measure if this execution has taken place. Several studies reported different follow-up durations ^{7, 30, 35} . Our findings are too inconsistent in reporting the timeframe to make any generalizations about the appropriate time to move intention to action; the literature on behaviour implementation suggests that this is an important avenue for future study.

The range of potential interceding factors that emerged in the review point to the fact that contraceptive behaviour is a complex psychosocial process shaped by the confluence of individual and contextual factors ¹⁰ . Such factors may help explain how pregnancy and relationship status are related to intentions or use of specific methods, whether goal or implementation intentions result in actual use, and over what timeframe intentions to use contraception are likely to transform into action. In turn, this can contribute to better understand people’s needs and preferences and how we can align programs to support them to achieve their reproductive goals and contraceptive goals.

There are several limitations to this review. There were relatively few studies that met the inclusion criteria. The relationship between ITU and contraceptive uptake was not the primary outcome of interest for those included papers. Thus, we had to calculate an odds ratio to estimate that relationship. Therefore, we treat our results as indicative. Another limitation is that the samples recruited for the included studies were primarily pregnant or postpartum samples—the desire to start sexual activity and contraception may be different for these populations compared to others. Geographic settings, particularly the difference in health systems and contraceptive access, may also explain the differences we found. In addition, other factors (e.g., cultural and social norms, knowledge about contraceptive methods, personal beliefs) may all contribute to reproductive and contraceptive intentions, decision-making, and subsequent use, and require further consideration.