The Methodological Pitfalls Complicating the Analysis of Early Term Births Authors: Ye'elah E. Berman, John P. Newnham, Dorota A. Doherty, Sarah V. Ward Published: 2026 ABSTRACT ABSTRACTBackgroundEvaluating the success of policies aiming to reduce early term birth (37+0–38+6 weeks' gestation) is analytically challenging.ObjectivesDetermine the best method of reporting temporal changes of early term birth rates.MethodsWe analysed 54,646 singleton births at Western Australia's main tertiary hospital (2009–2019). Using interrupted time series, segmented and nominal regression, we assessed how varying gestational age outcomes, denominator, or statistical method affected conclusions regarding temporal changes (pre and post‐Initiative implementation) in the timing of birth.ResultsInterrupted time series showed yearly rates of birth at 37+0–38+6 weeks increased pre‐Initiative, with no instant level change or change in the slope post‐Initiative, regardless of denominator: all births (slope change −0.0, 95% CI −0.5, 0.5); births > 36+6 weeks (slope change: −0.4, 95% CI −1.0, 0.2), and no change to the number of early term births averted. Conversely, when examining births < 39+0 weeks, pre‐Initiative rates increased by 0.9% (95% CI 0.5, 1.3) per year, then stabilised post‐Initiative (slope change: −0.8, 95% CI −1.4, −0.3), with no change to the number of births < 39+0 weeks averted. Nominal logistic regression showed that pre‐Initiative, the rise in the early term birth rate was driven by a reduction in births at ≥ 39+0 weeks, and post‐Initiative by a reduction in preterm birth. Using segmented quantile regression, lengthening of gestational age at birth in days was observed more so at the lower percentiles, compared to the estimated counterfactual.ConclusionExamining early term birth rates in isolation can mask changes in birth timing in the presence of a changing preterm or ≥ 39+0 birth rate. Researchers should examine all births < 39+0 weeks rather than just early term births, or use nominal logistic regression. For gestational age in days as the outcome, segmented quantile regression can be used. Introduction Preterm birth, defined as birth < 37+0 completed weeks gestation, is the leading cause of death and disability in children less than 5 years of age globally [1, 2]. In response to increases in the preterm birth rate in the state [3], the Western Australian (WA) Preterm Birth Prevention Initiative (the Initiative) was launched in 2014 [4], with the primary aim of reducing preterm birth. In the first year following the introduction of the Initiative, the preterm birth rate was lowered across the state [4], and 5 years later it was still decreasing amongst non‐Aboriginal women at the tertiary hospital [5]. Timing of birth is a spectrum, and infants born just beyond the preterm period share many features with those born preterm [6]. Compared to birth at 39+0–40+6 weeks, birth at 37+0–38+6 weeks is associated with higher risks of morbidity [7] and mortality [8] during the perinatal period, as well as adverse outcomes during childhood [9, 10] and adulthood [11]. In light of these differences, the American College of Obstetricians and Gynecologists recommend that “term” gestational ages be further broken down into early term (37+0–38+6), full term (39+0–40+6), and late term (41+0–41+6) [12, 13]. Due to the higher risks associated with early term birth, a secondary aim of the Initiative was to lower the rate of births between 37+0 and 38+6 weeks. We have previously published on assessing the efficacy of the Initiative in reducing the rates of preterm birth [4, 5], with further studies assessing reductions in the rate of early term birth ongoing. However, in evaluating the success of this policy, we faced complex analytical challenges due to the choice of denominator, method of statistical analysis and assessment of timing of birth in gestational weeks or days, plus boundary effects caused by categorisation of the continuous gestational age outcome. As such, this paper aimed to assess the effect that varying these parameters had on conclusions regarding changes in the early term birth rates. We hypothesised that different analytical methods and outcome categorisations would result in different conclusions about temporal changes in early term birth rates. Methods Study Population and Data Sources This study used all singleton births to non‐Aboriginal women, who gave birth at the main tertiary hospital, King Edward Memorial Hospital, in WA between 2009 and 2019. Pregnancy data were obtained from the Midwives Notification System, a register of all births in WA attended by a midwife and occurring at ≥ 20 weeks gestation, or with a birthweight of ≥ 400 g. Outcomes The outcome for this study was gestational age at birth. For interrupted time series analyses, we examined rates of birth at < 37+0 weeks gestation (preterm), 37+0–38+6 weeks gestation (early term) and < 39+0 weeks gestation (early birth). For segmented nominal logistic regression, we assessed all births, categorised into < 37+0 weeks gestation, 37+0–38+6 weeks and ≥ 39+0 weeks. We also assessed changes in the gestational age at birth in days. Exposure The primary exposure was time‐period, contrasting birth rates from before (2009–2014) and after (2014–2019) implementation of the Initiative in May 2014. We manipulated the outcome category (< 37+0 weeks vs. 37+0–38+6 weeks vs. < 39+0 weeks vs. gestational age in days), the choice of denominator (all births vs. births > 36+6 weeks), and the type of analysis employed (interrupted time series vs. segmented nominal logistic or quantile regression) and assessed the effect this had on our conclusions regarding changes in the rate of birth at early term associated with the Initiative. Statistical Analysis Frequency distributions for demographic characteristics, maternal conditions and obstetric history were reported by time‐period (pre‐ and post‐Initiative). Interrupted time series analyses estimated changes in the trends and levels of the proportion of births that occurred at < 37+0, 37+0–38+6 and < 39+0 weeks gestation, following implementation of the Initiative. By extending the trend line from the pre‐Initiative time‐period, we created the counterfactual estimated trend line, representing the estimated number of births to occur at these gestations in the absence of the Initiative. The counterfactual was then compared to the line of the best fit that was plotted based on the actual births occurring at these gestations post‐Initiative. Multiplying the difference between the two lines by the actual number of births per month gave the estimated number of births averted in each gestational age group in the post‐Initiative period, with 95% confidence intervals (CI) derived using bootstrapping methods with 1000 iterations. Separate slopes (the average yearly change in the birth rate) for both the pre‐Initiative and post‐Initiative time‐periods were also derived. The percentage changes between the pre‐ and post‐Initiative slopes and one reflecting the immediate effect of the introduction of the Initiative (at the breakpoint in May 2014) were also calculated. For each analysis, we used a different combination of gestational age outcome and denominator, contrasted the resulting slope and change coefficients, and reported the number of births averted at that gestation. Unadjusted segmented nominal logistic regression with time in months, intervention (pre‐ vs. post‐Initiative) and a time*intervention interaction was used to simultaneously model the predicted probabilities of birth at < 37+0, 37+0–38+6 and ≥ 39+0 weeks' gestation. Yearly changes in the log odds of preterm birth (< 37+0) and early term birth at 37+0–38+6 weeks, compared to birth at ≥ 39+0 weeks, were reported for the pre‐ and post‐Initiative time‐periods, as well as the level change in log odds at the time of the Initiative. Births averted at each gestation over the post‐Initiative period were calculated using the same methods as the interrupted time series analyses. Segmented quantile regression, as described by Moineddin and colleagues in 2022 [14], was performed using the SAS function PROC QUANTREG, with the outcome as gestational age in days and covariates for time, intervention, and a time*intervention interaction included [14]. This methodology allowed us to plot the predicted values for the gestational length in days at the 25th, 40th, 60th, and 75th percentile of each month. These percentiles were chosen as they covered the gestational age range of interest. As in the previous analyses, the model yielded pre‐Initiative and post‐Initiative slopes representing the yearly change in gestational length in days at each quantile. We also reported a level change at the time of the Initiative, a slope change between the pre‐ and post‐Initiative slopes, and the difference in predicted gestational length between the actual values post‐Initiative and those predicted by the counterfactual line, averaged across the post Initiative time‐period. SAS (version 9.4, Cary, NC: SAS Institute Inc.) statistical software package was used for data analyses. Missing Data and Sensitivity Analysis The Index of Relative Socio‐Economic Advantage and Disadvantage (IRSAD) score was missing for 993 (1.8%) births categorised as ‘unknown’ for reporting. There were 9227 (16.9%) births to multiparous women for whom a record of their previous pregnancy was not available. These women were classified as having an unknown history of preterm birth and no prior stillbirth. All other data were complete. Only monthly rates are reported, as sensitivity analyses indicated that these produced similar results to yearly rates. Results There were 64,993 births at the state's main tertiary hospital between 2009 and 2019, 55,065 of which were singleton births to non‐Aboriginal women. A further 419 births were excluded as terminations of pregnancy, and the final sample size consisted of 54,646 births. Table 1 shows the number and percentage of births in the study population by year. The rate of preterm birth (< 37+0 weeks gestation) increased from 18.5% in 2009 to a maximum of 20.1% in 2013, before reducing to a minimum of 14.7% in 2019. The rate of early term birth (37+0–38+6 weeks' gestation) was lowest in 2009 (27.7%) and rose steadily to a maximum of 33.3% in 2018. Births that were not considered early (≥ 39+0 weeks) reduced in the pre‐Initiative period (53.8% in 2009 to 48.9% in 2013) and then stabilised in the post Initiative period (50.7% in 2014 to 52.5% in 2019). Women who gave birth were slightly older and had slightly higher rates of pre‐existing diabetes in the post‐Initiative time‐period, compared to the pre‐Initiative time‐period (Table 2). Post‐Initiative, they also had lower rates of asthma and other pre‐existing conditions, fewer previous children, and substantially lower rates of smoking. TABLE 1The number and percentage of singleton births per year to non‐Aboriginal women at King Edward Memorial Hospital (2009–2019).YearAll births< 37+0 weeks37+0–38+6 weeks≥ 39+0 weeks N = 54,646 n = 9410% n = 16,810% n = 28,426%2009487390218.5134827.7262353.82010498183916.8149630.0264653.12011491084217.1138928.3267954.62012519993117.9147928.4278953.62013499799820.1155531.1244448.92014505194518.7154730.6255950.72015492676315.5153031.1263353.52016489182316.8160932.9245950.32017503684316.7162332.2257051.02018481479516.5160133.3241850.22019496872914.7163332.9260652.5 The number and percentage of singleton births per year to non‐Aboriginal women at King Edward Memorial Hospital (2009–2019). TABLE 2Frequencies and percentages of maternal characteristics by time‐period (pre‐ vs. post‐Initiative), for non‐Aboriginal women giving birth to singletons at King Edward Memorial Hospital (2009–2019).Total (n = 54,646)Pre initiative (n = 26,633)Post initiative (n = 28,013)January 2009–December 2019January 2009–April 2014May 2014–December 2019 N n % n %Maternal demographicsMaternal age< 2014999643.65351.920–3440,29819,83774.520,46173.0≥ 3512,849583221.9701725.0ParityNulliparous24,19011,55443.412,63645.11–429,40214,48054.414,92253.3≥ 510545992.34551.6Smoked during pregnancyNo48,92423,14986.925,77592.0Yes5722348413.122388.0Index of relative socio‐economic advantage and disadvantage (IRSAD) in lowest 20% a No48,60624,13890.624,46887.4Yes504723648.926839.6Unknown9931310.58623.1Maternal conditionsPre‐existing diabetesNo53,24226,07397.927,16997.0Yes14045602.18443.0Pre‐existing hypertensionNo53,69026,19898.427,49298.1Yes9564351.65211.9AsthmaNo49,44623,79589.325,65191.6Yes5200283810.723628.4Other pre‐existing conditions b No31,90713,03749.018,87067.4Yes22,73913,59651.1914332.6Obstetric historyPrevious preterm birthNo16,971774029.1923133.0Yes425820807.821787.8Unknown33,41716,81363.116,60459.3Previous stillbirthNo53,09725,87297.127,22597.2Yes15497612.97882.8Caesarean last deliveryNo44,72121,87982.222,84281.5Yes9925475417.9517118.5In Vitro fertilisation conceptionNo53,27026,05397.827,21797.2Yes13765802.27962.8 a Socioeconomic status was assigned based on the maternal residence postcode recorded on the MNS, using the Index of Relative Socio‐Economic Advantage and Disadvantage (IRSAD) score, derived by the Australian Bureau of Statistics [15]. b Other Pre‐Existing Conditions include any conditions diagnosed before pregnancy, which influence care during the pregnancy. Instructions for coding the ‘other’ field can be found at https://www.health.wa.gov.au/~/media/Files/Corporate/general‐documents/Data‐collection/PDF/Guidelines‐for‐Completion‐of‐NOCA‐2023.pdf. Frequencies and percentages of maternal characteristics by time‐period (pre‐ vs. post‐Initiative), for non‐Aboriginal women giving birth to singletons at King Edward Memorial Hospital (2009–2019). Socioeconomic status was assigned based on the maternal residence postcode recorded on the MNS, using the Index of Relative Socio‐Economic Advantage and Disadvantage (IRSAD) score, derived by the Australian Bureau of Statistics [15]. Other Pre‐Existing Conditions include any conditions diagnosed before pregnancy, which influence care during the pregnancy. Instructions for coding the ‘other’ field can be found at https://www.health.wa.gov.au/~/media/Files/Corporate/general‐documents/Data‐collection/PDF/Guidelines‐for‐Completion‐of‐NOCA‐2023.pdf. Birth Rates by Completed Weeks Gestation—Interrupted Time Series Analyses Figure 1A shows the rates of birth at 37+0–38+6 weeks from 2009 to 2019, with births > 36+6 weeks gestation in the denominator. In the pre‐Initiative period (2009–2014), the rate of early term birth was increasing each year. After the introduction of the Initiative, there was no level change or notable change in the slope. Compared to the counterfactual, the Initiative was not associated with a reduction in early term births across the post‐Initiative time‐period. Monthly percentages of singleton births to non‐Aboriginal women, with predicted probabilities from interrupted time series analysis, at King Edward Memorial Hospital from 2009 to 2019. (A) includes births between 37+0 and 38+6 weeks in the numerator, with all births > 36+6 weeks in the denominator. (B) includes births between 37+0 and 38+6 weeks in the numerator, with all births in the denominator. (C) includes births < 39+0 completed weeks in the numerator, with all births in the denominator. Blue dots represent the monthly percentage of births at the given gestational that occurred in the pre‐Initiative time period (January 2009–May 2014), with the solid blue line representing the trend line generated by the regression model. The orange dots and line represent the monthly percentage and trend line for the post‐Initiative time period (May 2014–December 2019). The dashed blue line represents the counterfactual predicted trend line that may have occurred in the absence of the Initiative. Rates of birth at 37+0–38+6 weeks with all births in the denominator are shown in Figure 1B. Similar overall trends were observed to those with births > 36+6 weeks as the denominator. During the pre‐Initiative period, the rate of early term birth was increasing each year. After the introduction of the Initiative, there was no instant level change or change in the slope observed. There was no difference between the actual and counterfactual regression lines for births at 37+0–38+6 weeks across the post‐Initiative time‐period. In comparison, Figure 1C shows that the rate of early births (< 39+0 weeks) was increasing rapidly each year during the pre‐Initiative period and the rate then stabilised post‐Initiative. No acute level change was observed at the time of the Initiative's implementation and despite the change in slope, the difference between the actual and the counterfactual regression lines did not quite reach a change in the number of births at < 39+0 weeks post‐Initiative. Supplementary Figure S1 shows that the rate of preterm birth (< 37+0 weeks) was increasing during the pre‐Initiative period, and then following a slope change, flattened in the post‐Initiative time period. This equated to a large number of preterm births averted post‐Initiative. Table 3 summarises pre and post‐Initiative slopes, level changes at the time of the Initiative, and births averted at the gestation of interest, for each analysis in this paper. TABLE 3Comparison of births averted at the gestation of interest, pre and post‐Initiative slopes, slope change, level change and conclusion by outcome, denominator and statistical analysis method used for births to non‐Aboriginal women at King Edward Memorial Hospital (2009–2019).Analysis typeOutcome categoryBirths in denominatorBirths averted/change in average gestational length in daysPre‐Initiative slope (2009–2014)Post‐Initiative slope (2014–2019)Slope changeLevel change at the time of the InitiativeConclusion about changes associated with the InitiativeInterrupted time series (slope: yearly change in proportion of births at the gestational age group)Percentage of births at 37+0–38+6 > 36+6 weeks266 (−729, 1239)0.8 (0.3, 1.2)0.4 (−0.6, 1.3)−0.4 (−0.1, 0.2)−0.0 (−1.8, 1.8)→ no effectPercentage of births at 37+0–38+6 All−249 (−751, 1241)0.4 (0.1, 0.8)0.4 (−0.4, 1.3)−0.0 (−0.5, 0.5)0.8 (−0.7, 2.3)→ no effectPercentage of births at < 39+0 All997 (−43, 2075)0.9 (0.5, 1.3)0.1 (−0.8, 1.0)−0.8 (−1.4, −0.3)−1.4 (−3.1, 0.3)↓ benefit (slope change, but not births averted)Percentage of births at < 37+0 All1246 (366, 2112)0.5 (0.2, 0.8)−0.4 (−1.1, 0.3)−0.8 (−1.2, −0.4)−2.2 (−3.5, −0.9)↓ benefit (slope change and births averted)Segmented nominal logistic regression (slope: yearly change in log odds compared to giving birth at ≥ 39 + 0 weeks)Percentage of births at ≥ 39+0 All−1012 (−965, 3032)refrefrefrefn/aPercentage of births at 37+0–38+6 All−265 (−797, 1308)0.0 (0.0, 0.1)0.0 (−0.0, 0.0)−0.0 (−0.0, −0.0)0.0 (−0.1, 0.1)→ no effectPercentage of births at < 37+0 All1278 (342, 2234)0.00 (0.0, 0.1)−0.0 (−0.1, 0.0)−0.1 (−0.1, −0.0)−0.1 (−0.2, −0.1)↓ benefit (slope change, and births averted)Segmented quantile regression (slope: average change in gestational age in days per year)Gestational length in days at the 25th percentile of each monthAll2.7 (0.4, 5.1)−0.5 (−0.7, −0.2)0.2 (−0.3, 0.6)0.6 (0.3, 1.0)1.0 1 (−0.4, 2.4)↓ benefit (slope change, and change in days)Gestational length in days at the 40th percentile of each monthAll1.0 (−1.0, 2.6)−0.3 (−0.5, −0.1)0 (−0.3, 0.3)0.3 (0.1, 0.6)0.6 (−0.2, 1.3)↓ benefit (slope change, not change in days)Gestational length in days at the 60th percentile of each monthAll0.7 (−1.5, 2.4)−0.3 (−0.4, −0.2)−0.2 (−0.4, −0.0)0.1 (−0.0, 0.3)0.3 (−0.5, 1.2)→ no effectGestational length in days at the 75th percentile of each monthAll0.7 (−0.5, 2.1)−0.3 (−0.4, −0.2)0.0 (−0.2, 0.2)0.3 (0.2, 0.4)−0.6 (−1.6, 0.4)↓ benefit (slope change, not change in days) Comparison of births averted at the gestation of interest, pre and post‐Initiative slopes, slope change, level change and conclusion by outcome, denominator and statistical analysis method used for births to non‐Aboriginal women at King Edward Memorial Hospital (2009–2019). Birth Rates by Completed Weeks Gestation—Segmented Nominal Logistic Regression Actual monthly birth rates (2009–2019), as well as predicted probabilities derived from the nominal logistic regression analysis investigating associations between year and month of birth, and gestational age at birth are shown in Figure 2. Pre‐Initiative, the log odds of having a preterm birth compared to giving birth at ≥ 39+0 weeks were increasing each year. Following the Initiative's implementation, there was an immediate reduction and a flattening of the slope. Monthly percentage of singleton births (dots) to non‐Aboriginal women at < 37+0, 37+0–38+6 and ≥ 39+0 weeks, with predicted probabilities (solid lines) for the pre‐Initiative (January 2009–May 2014) and post‐Initiative (May 2014–December 2019) time periods, and the prediction in the absence of the Initiative (dashed line), derived by segmented nominal logistic regression analysis (King Edward Memorial Hospital, 2009–2019). For early term births, the log odds compared to birth at ≥ 39+0 weeks were also increasing each year prior to the Initiative, with no level or slope change observed in the post‐Initiative period. Over the 5 years of the Initiative, the changes in timing of birth equated to a reduction in the number of preterm births, but no change in the number of early term births or the number of births at ≥ 39+0 weeks. Mean Gestational Age in Days—Segmented Quantile Regression The actual and predicted monthly values for gestational age at birth in days, for babies born at the 25th, 40th, 60th and 75th percentiles (2009–2019) are shown in Figure 3. Predicted values were derived from coefficients generated by segmented quantile regression. Pre‐Initiative, births at all calculated quantiles were getting earlier (Table 3), with births at the 25th percentile (which fell within the 37 weeks gestational age range) decreasing at the fastest rate. There was no instant level change at any percentile, but there were changes in slope, with gestations becoming longer in the post‐Initiative time period compared to the pre‐Initiative time period at the 25th, 40th, and 75th percentiles. Across the post‐Initiative time period, the average gestational length in days at the 25th percentile was longer than the prediction based on the counterfactual line. There were no differences at the 40th, 60th or 75th percentiles. Prediction for the monthly (dots) gestational age (in days) at birth for the 25th, 40th, 60th and 75th percentile of singleton births to non‐Aboriginal women at King Edward Memorial Hospital from 2009 to 2019. Trend lines (solid lines) and counterfactual predictions (dashed lines) for the pre‐Initiative (January 2009–May 2014) and post‐Initiative (May 2014–December 2019) periods were ascertained using segmented quantile regression. Comment Principal Findings The results of this paper demonstrate that when assessing temporal changes in the timing of birth at early term, the categorisation of the outcome and the method of analysis used can affect interpretation of changes associated with interventions to reduce early birth. When modelling gestational age at early term alone, our results showed a year‐on‐year increase in the proportion of births at 37+0–38+6 weeks, with negligible change in trajectory resulting from the Initiative. Consequently, no early‐term births were averted. Analysis of birth rates at < 39+0 and < 37+0 weeks showed that in fact the composition of early term births had changed substantially. Segmented nominal logistic and quantile regression were proposed as alternatives to interrupted time series analysis. Strengths of the Study Our study included 11 years of data covering all births at WA's major tertiary hospital, providing a large sample size and ample statistical power to perform robust time‐series analyses. Data were drawn from the Midwives Notification System, which is a validated register of all births in WA that are attended by a midwife, ensuring near complete case ascertainment. The study is also directly policy relevant, using the example of the Initiative to demonstrate how methodological choices can alter conclusions about policy effectiveness. Limitations of the Data Our study was limited by the lack of concurrent controls, but given we were working with administrative data, our study design was appropriate. We also only used data from King Edward Memorial Hospital as the preterm birth rate had decreased substantially at this site, and consequently it made an excellent case study for demonstrating how examining the early term birth rate in isolation can be misleading in the presence of a changing preterm birth rate. As King Edward Memorial Hospital is the major tertiary hospital in WA, it receives a significant number of high‐risk cases from across the whole state. As such, it has a much higher preterm birth rate than both the state as a whole and the national average, and is not representative of the state's preterm birth figures. As our study used administrative data, we do not have specific information on what method was used for dating of gestational age, but the gestational age in weeks variable used was calculated based on a validated algorithm using a combination of last menstrual period, expected due date and date of birth [16]. We also did not perform adjusted analyses, despite having identified several risk factors for early birth (including smoking, maternal age, diabetes, asthma and parity) that were different between the pre‐ and post‐Initiative periods. As the purpose of this paper was to demonstrate different methods for examining temporal changes in early birth, and not to evaluate the effectiveness of the WA Preterm Birth Prevention Initiative, they were not considered relevant to include here. Adjusted analyses examining the Initiative's effectiveness have been published elsewhere [4, 5, 8, 17]. Interpretation In reporting results from only early term (37+0–38+6 weeks) and using interrupted time series analyses, an important shift in the timing of birth was obscured. Analyses of births at 37+0–38+6 weeks (regardless of denominator) showed no notable effect on the rate of births at early term following implementation of the Initiative. In contrast, examination of the rates of births at < 37+0 and < 39+0 weeks' gestation showed that prior to the Initiative, rates of both preterm birth and early birth at < 39+0 weeks were increasing each year. Then, following the Initiative, there was a change in trajectory and the change in these rates plateaued, with 1246 preterm births averted across the post‐Initiative time period. From these results, it is clear that the composition of the early term birth rate had changed substantially. Where previously the increase in early term birth was driven entirely by a reduction in births ≥ 39+0 weeks, after the Initiative, the increase in the early term birth rate was driven by the reduced preterm birth rate, as gestation length was increased and pushed into the early term timeframe. These results make it clear that it is not appropriate to assess whether interventions aiming to change the timing of birth have been effective by using early term birth alone as an outcome. Analysing births < 39+0 weeks cannot directly answer questions about associations between the introduction of an intervention and the rate of early term birth because the preterm births are also included in this categorisation. However, it did show us that there was a flattening in the year‐on‐year increase in the rate of overall birth < 39+0 weeks, which is of significant benefit. The analytical problem we have identified is generalisable to any continuous outcome that has been categorised and has a lower and upper bound, indicating that a different method of analysis, such as nominal logistic regression, must be employed. Nominal logistic regression models are a preferable method of analysis as they have the advantage of being able to simultaneously model all three gestational age categories, and consequently the previously observed shifts in gestational age are apparent. However, while predicted probabilities can provide a convenient visual description of temporal changes in the timing of birth, only changes in log odds associated with the Initiative (or any intervention) can be reported. This makes year‐on‐year changes between time periods difficult to interpret, as they are in relation to the reference category (births ≥ 39+0 weeks gestation) rather than stand‐alone rates that can be reported as in an interrupted time series analysis. Probabilities can still be compared to the counterfactual predictions, and when combined with the actual birth rates that occurred, the estimated number of preterm and early term births averted can still be calculated. The results of the segmented nominal logistic regression were similar to those from the four interrupted time series analyses, showing many preterm births averted and no change in early term births. However, the use of nominal logistic regression makes it clear that the composition of early term births was entirely different between the pre‐ and post‐Initiative time periods and that despite there being no change in the early term birth rate, the Initiative did have an effect on the timing of birth. Analysis of gestational age in days is similarly complicated. One cannot analyse the average gestation at birth across the whole gestational age range because interventions aimed at optimising perinatal outcomes seek to increase short gestational ages to as close to ≥ 39+0 weeks as possible, but do not seek to lengthen longer gestations. Analysing the average gestational age at < 39+0 weeks or at 37+0–38+6 weeks is also not an adequate solution because these populations are not fixed. A reduction in births ≥ 39+0 weeks can result in an artificial increase in the mean gestational age in days, while a reduction in the preterm birth rate can result in an artificial decrease. The use of segmented quantile regression can circumvent some of these problems. As this method doesn't require categorisation of gestational age, it allows us to model changes at a range of quantiles so that it is apparent if the intervention of interest affects different parts of the gestational age range differently. Our results showed that the trajectories of the gestational age in days in all calculated percentiles were getting shorter in the pre‐Initiative time period, with this trajectory changing in the post‐Initiative time period. At the 60th and 75th quantiles, which both fell above 39+0 weeks, the difference between the actual and counterfactual lines was smaller than at the 25th and 40th percentiles, and at the 60th percentile there was no change in the slope. This makes sense because once a pregnancy reaches 39+0 weeks (considered full term), attempts would not be made to delay delivery if spontaneous labour occurs or complications necessitating delivery arose. The only percentile in which there was a reduction in gestational age in days across the post‐Initiative period was the 25th percentile, falling within the 37 week gestational age period. Conclusions When assessing changes in the timing of early term birth, careful consideration must be given to the impact of denominators and boundaries imposed on continuous outcomes, as they can result in misleading conclusions, obscuring important results. Temporal changes in rates of birth at early term cannot be accurately commented on without considering the broader context in analyses. For analyses using completed weeks gestation as the outcome, we recommend either analysing all early births < 39+0 weeks instead of early term births. If early term births are of particular interest, we recommend using nominal logistic regression to simultaneously model probabilities of preterm birth, early term birth, and birth ≥ 39+0 weeks. If the outcome is gestational age at birth in days, segmented quantile regression can show how interventions can have differential effects at different gestational ages. We are unaware of any other papers that have investigated this issue. This paper makes an important contribution to the methodological literature for those attempting to assess temporal changes in the timing of birth. Author Contributions Y.E.B. conceived of the project, designed the work, analysed and interpreted the data and drafted the manuscript. J.P.N. interpreted the data and edited and revised the manuscript. D.A.D. edited and revised the manuscript. S.V.W. interpreted the data and edited and revised the manuscript. Funding This study was supported by the National Health and Medical Research Council (Partnership Grant APP1151853 www.nhmrc.gov.au), Commonwealth Department of Health and Ageing grant (grant ref.: GA217476), the McCusker charitable foundation, and Channel 7 Telethon Trust (www.Telethon7.com). None of the study sponsors had any role in the design, data collection, analysis, interpretation of data, writing of the manuscript or decision to submit the paper for publication. Disclosure This article does not make use of previously published material. Ethics Statement The study was approved by the Women and Newborn Health Service Human Research Ethics Committee (RGS0000002677), the Health Department of Western Australia Human Research Ethics Committee (RGS0000000704), and the Western Australian Aboriginal Health Ethics Committee (965). The study met the criteria for a waiver of consent outlined in section 2.3 of the National Statement on Ethical Conduct in Human Research, and this was granted by the Health Department of Western Australia Human Research Ethics Committee. All methods were carried out in accordance with the Western Australian Practice Code for the Use of Personal Health Information, and counts less than 5 were suppressed. Consent A waiver of consent was obtained for this study. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: Monthly percentages of singleton early births at ≤ 37+0 weeks, to non‐Aboriginal women, at King Edward Memorial Hospital from 2009 to 2019. Blue dots represent the monthly percentage of births at ≤ 37+0 weeks gestation that occurred in the pre‐Initiative time period (January 2009–May 2014), with the solid blue line representing the trend line generated by the regression model. The orange dots and line represent the monthly percentage and trend line for the post‐Initiative time period (May 2014–December 2019). The dashed blue line represents the counterfactual predicted trend line that may have occurred in the absence of the Initiative.