Determinants of length of stay after cesarean sections in the Friuli Venezia Giulia Region (North-Eastern Italy), 2005–2015

Since Italy has the highest cesarean section (CS) rate (38.1%) among all European countries, the containment of health care costs associated with CS is needed, along with control of length of hospital stay (LOS) following CS. This population based cross-sectional study aims to investigate LoS post CS (overall CS, OCS; planned CS, PCS; urgent/emergency CS, UCS), in Friuli Venezia Giulia (a region of North-Eastern Italy) during 2005–2015, adjusting for a considerable number factors, including various obstetric conditions/complications. Maternal and newborn characteristics (health care setting and timeframe; maternal health factors; child’s size factors; child’s fragility factors; socio-demographic background; obstetric history; obstetric conditions) were used as independent variables. LoS (post OCS, PCS, UCS) was the outcome measure. The statistical analysis was conducted with multivariable linear (LoS expressed as adjusted mean, in days) as well as logistic (adjusted proportion of LoS > 4 days vs. LoS ≤ 4 days, using a 4 day cutoff for early discharge, ED) regression. An important decreasing trend over time in mean LoS and LoS > ED was observed for both PCS and UCS. LoS post CS was shorter with parity and history of CS, whereas it was longer among non-EU mothers. Several obstetric conditions/complications were associated with extended LoS. Whilst eclampsia/pre-eclampsia and preterm gestations (33–36 weeks) were predominantly associated with longer LoS post UCS, for PCS LoS was significantly longer with birthweight 2.0–2.5 kg, multiple birth and increasing maternal age. Strong significant inter-hospital variation remained after adjustment for the major clinical conditions. This study shows that routinely collected administrative data provide useful information for health planning and monitoring, identifying inter-hospital differences that could be targeted by policy interventions aimed at improving the efficiency of obstetric care. The important decreasing trend over time of LoS post CS, coupled with the impact of some socio-demographic and obstetric history factors on LoS, seemingly suggests a positive approach of health care providers of FVG in decision making on hospitalization length post CS. However, the significant role of several obstetric conditions did not influence hospital variation. Inter-hospital variations of LoS could depend on a number of factors, including the capacity to discharge patients into the surrounding non-acute facilities. Further studies are warranted to ascertain whether LoS can be attributed to hospital efficiency rather than the characteristics of the hospital catchment area.

The database. Data from the 11 maternity services of FVG during calendar years 2005-2015 were extracted from the Regional Repository, an electronic database anonymously storing administrative information from the Italian NHS. The database we analyzed included information from two sources: the hospital discharge forms (HDF, using the respective ICD-9 codes) and the Certificate of Delivery Care (CEDAP, Italian acronym), a formatted questionnaire collecting clinical and personal information on women and newborns (supplementary material, S1) 8,9,13,14,29,30 .
We used the following ICD-9 codes to retrieve the obstetric conditions associated with each childbirth: The rest of data derived from CEDAP, in which delivery mode is defined as follows: 1. Vaginal delivery (VD) without forceps or vacuum extraction; 2. Planned CS (PCS) or CS for failed induction; 3. CS during labour or urgent CS; 4. Forceps extraction; 5. Vacuum extraction; 6. Other forms of VD.
For the purpose of this study, we used the categories 2 and 3, incorporated into OCS. Category 3 indicates UCS.
The 11 facility centres of FVG were anonymized and coded by alphabetic letter from A to K. A and B are second level maternity units (> 1000 annual births and equipped with a neonatal intensive care unit), whereas the other 9 are first level (< 1000 annual births and/or devoid a neonatal intensive care unit). Figure 1 shows the flowchart displaying the various criteria applied to the initial database to obtain the final number of hospital births available for the analysis 8 .
Length of hospital stay after childbirth. LoS (measured in number of whole days) was calculated by subtracting the date of birth by CS from the date of hospital discharge.
Statistical analysis. The mean LoS and the percentage of LoS longer than the proposed ED benchmark following CS (4 days) were calculated for each of the above explanatory factors. The mean LoS and the 0/1 variable LoS (lower/higher than ED) were used as outcomes in a multiple logistic and in a multiple linear regression models, respectively (see below). Some factors were deliberately dropped from the final multivariate logistic and linear regression model for the following different reasons: • Apgar score at 1 min and resuscitation due to collinearity with Apgar score at 5 min and intensive care unit (ICU) admission respectively, which both had stronger effect size and we thought they were more plausible to be retained in the final model;  Stepwise backward selection of independent variables was used to build up all final linear regression models, using p < 0.05 as a criterion. Results were expressed as regression coefficient (RC) with 95% confidence interval (95%CI) and reported in two tables: factors related to mother and newborn health (Fig. 3a); and hospital comparison (Fig. 3b).
Additionally, we fitted a multiple logistic regression model for each CS (OCS, PCS as well as UCS), using LoS as a binary outcome (LoS > ED vs. LoS ≤ ED). Results were expressed as odds ratio (OR) with 95% confidence interval (95%CI) and reported in two tables (Fig. 4a,b) as above.   www.nature.com/scientificreports/ Results of all regression models (logistic as well as linear) were obtained by comparing each stratum specific estimate (OR and RC) with the corresponding reference category. Hospital J was chosen as reference among all maternity centres, since it was the third maternity centre of FVG in terms of yearly number of births during the entire study period, had the shortest mean LoS after CS among all public hospitals and the second highest CS rate in the region.
Considering the large number of statistical tests performed in the multivariable regression models, some p-values could have been significant by chance. Therefore, we employed as a further selection approach the procedure proposed by Benjamini-Hochberg (BH), setting the false discovery rate at 5% to obtain the BH p-value to be associated with each risk estimate 34 .
Missing values were excluded and complete case analysis was performed. Stata 14.2 (College Station, Texas, USA) was employed for the analysis.

LoS (days) M ± SD > 4 (%) M ± SD > 4 (%) M ± SD > 4 (%)
Child's size factors    www.nature.com/scientificreports/ variation in the mean LoS after CS, which for PCS varied from 4.0 days in centre J up to 5.3 days in D. The proportion of LoS > ED ranged from 11.7% in centre C to 85.4% in D. Considering UCS, the mean LoS after CS varied from 4.3 days in centre C up to 5.6 days in D. The proportion of LoS > ED for UCS ranged from 15.5% in centre C up to 88.5% in facility D. Examining each hospital separately, the estimates were higher after UCS than PCS, regardless LoS was expressed as mean or percentage > ED. Lastly, increasing mean LoS post CS and LoS > ED were found with higher number of hospital admissions and number of births on delivery day. Always of note from Table 2, for any DM the mean LoS as well as the rates of LoS > ED were smaller on Mondays and Tuesdays, whereas they were higher during winter (December-February) and Spring (March-May) months. As can be seen from Table 2, across all CS types the mean LoS and the proportion of LoS > ED was particularly higher among women affected by hypertension/diabetes or admitted earlier to hospital (pre-delivery LoS 6+ days). Mean Los and LoS > ED were instead greater among mothers of older age (> 45 years) following PCS, and among stillbirths for UCS. As reported above, larger variations in the outcome measures were found for UCS as compared to PCS and for LoS > ES with respect to mean LoS. Table 3 shows the distribution of the mean LoS and the proportion of LoS > ED after CS by clinical factors of the child. For all types of CS, both latter outcomes tended to increase with decreasing birthweight and child's size, with greatest estimates found for low birthweight [LBW (2.0-2.5Kg)]. By contrast, both mean LoS and LoS > ED consistently and considerably decreased as gestational age increased. Regarding placental weight, the outcome estimates were higher at the extremes of the distribution. Always form Table 3, the mean LoS and the proportion of LoS > ED post PCS and UCS tended to be higher in presence of child's fragility factors: Apgar score at 1 min < 7, Apgar score at 5 min < 8, ICU admission, resuscitation and multiple birth.
As can be seen from Table 4, for both CS types the mean LoS was rather consistent across socio-demographic factors, although it was higher among non-EU women and with lower level of education of both parents. LoS > ED increased progressively with paternal age for both PCS and UCS and was considerably higher among consanguineous parents after PCS. Table 5 shows the mean LoS and the proportion of LoS > ED after CS by obstetric history factors. Conflicting results between PCS and UCS were found by comparing the mean LoS with the proportion of LoS > ED for obstetric history factors. A reduction of LoS>ED was observed with increasing number of previous livebirths, history of CS and previous pre-term babies. Conversely, the mean LoS tended to increase with history of neonatal deaths and higher number of previous intentional abortions. Less important variations of both outcomes could be observed in relation with history of stillbirth, spontaneous abortions.    www.nature.com/scientificreports/ Table 6 shows the distribution of LoS by obstetric conditions/complications. It can be noted that the mean LoS and the proportion of LoS > ED were considerably greater for eclampsia/pre-eclampsia (after both PCS as well as UCS), placenta previa/abruptio placenta/ante-partum haemorrhage (after both PCS and UCS), shoulder presentation (more for UCS) and Rh iso-immunization (following UCS). The proportion of LoS > ED was considerably greater with UCS in case of no labour. For both CS types the mean LoS and the proportion of LoS > ED was remarkably higher for breech and shoulder presentation. Figure 3a displays the health factors of the mother and the newborn with significant effect on LoS (as a linear endpoint) following OCS, PCS and UCS. LoS significantly decreased over the years, with modest effect size but strong significance, particularly after PCS. Irrespective of the type of CS (PCS or UCS) a longer mean LoS was observed among non-EU women. By contrast, LoS had a clear inverse relation as the number of previous livebirths increased. LoS was also significantly shorter with history of CS only after PCS. An important increase in mean LoS was associated with eclampsia/pre-eclampsia, with much stronger significance following UCS. A greater LoS was observed in older women after PCS, with stronger and clearer trend as mother's age increased. Furthermore, a strongly significant increase of LoS was found at 33-36 weeks of gestation (irrespective of type of CS). Likewise, LoS was longer only in women with LBW infants (for both PCS and UCS). Multiple births involved    www.nature.com/scientificreports/ a significant increase in LoS following PCS. Moreover, regardless the type of CS, LoS was significantly longer when the mother was affected by hypertension/diabetes, particularly after PCS. Longer pre-delivery LoS as well as numerous obstetric checks during pregnancy had a tendency to increase LoS post-CS, mainly for UCS. Shoulder presentation resulted in longer LoS post UCS, whereas obstructed labour, placenta previa/abruptio placenta/ ante-partum haemorrhage and Apgar at 5 min score > 7 were factors associated with longer LoS mainly post PCS. Figure 3b relates to the differences among maternal centres of FVG. The RCs are expressed in the same unit (days) of the outcome variable (LoS). Both unadjusted and adjusted mean LoS was lower for PCS than UCS cases in all hospitals (Table 1 and Fig. 3b). All centres have a RC higher than the reference centre. Among PCS cases, RC was longer > 1 day in two centres (D, I), between 0.5 and 1 day in five hospitals (K, A, G, E, F in decreasing order of effect size), and < 0.5 day in two maternal units (B, C). Adjusted RCs were comparatively lower among UCS cases. Since they belong to the same model of multivariable linear regression, results of Fig. 3b are adjusted for the same factors displayed at the bottom of Fig. 3a. Therefore, the wide differences among maternity centres cannot be attributed to the case mix.

LoS (days) M ± SD > 4 (%) M ± SD > 4 (%) M ± SD > 4 (%)
Unlike Figs. 3a,b, 4a,b use LoS as binary outcome (LoS > ED vs. LoS ≤ ED) instead of linear endpoint. Since the same regression techniques were carried out in all tables, Fig. 4a,b were similar to Fig. 3a,b. Therefore, calendar year, number of previous livebirths and CS history were significantly associated with reduced odds of LoS > ED for both UCS and PCS (Fig. 4a). By contrast, for both CS types LoS > ED was more likely in non-EU mothers, eclampsia/pre-eclampsia, pre-term gestations (33-36 weeks), LBW (2.0-2.5 kg) and hypertension/diabetes. However, whilst the association of LoS > ED with pre-term gestation and with eclampsia/pre-eclampsia was much stronger for UCS, for LBW and mother's nationality it was stronger following PCS. Other important factors predominantly associated with LoS > ED after PCS were multiple birth and increasing maternal age (Fig. 4a). As can be seen from Fig. 4b, all maternity centres but C were by far more likely to surpass the ED benchmark than the reference (centre J). A similar pattern was observed in the multiple linear regression model (Fig. 3b), although in the latter model centre H was the maternity unit less differing from the reference. The discrepancy can be explained by the criterion "shortest mean LoS post CS among all public hospitals of FVG" used in the choice of hospital J as reference. Hospital C was the only private hospital in FVG.
Interestingly, as can be noted from Fig. 4a, the adjusted OR of LoS > ED was higher than reference (Monday) in all weekdays but Tuesday, with higher degree of significance for Wednesday and Thursday. Further, although with relatively weak significance, for all types of CS LoS > ED was significantly higher during spring months (March-May) than the reference (summer months, June-August). Figure 5a,b display the mean LoS and the proportion of LoS > ED over time in FVG, adjusted for the same factors included in the above mentioned linear (Fig. 3a,b) as well as logistic (Fig. 4a,b) regression models, respectively. As can be seen, there was a clear decreasing trend of LoS > ED over the years for all three types of CS, whilst the temporal diminishment of the mean LoS was less pronounced. Figure 6a,b display the mean LoS and the proportion of LoS > ED by maternity centres of FVG during the study period, adjusted for the same factors fitted in the above mentioned linear (Fig. 3a,b) as well as logistic (Fig. 4a,b) regression models. A clear adjusted hospital variability can be noted, more pronounced for LoS > ED.   May). With the exception of mother's nationality (very strong association), prolonged LoS was mainly driven by the clinical conditions of the mother (eclampsia/pre/eclampsia, hypertension/diabetes) and the newborn (gestational age < 36 weeks, birthweight 2.0-2.5Kg). After adjusting for the major medical and obstetric conditions/complications, the strongest determinant of LoS post CS was inter-hospital variation. All maternity centres but C were by far more likely to surpass the ED benchmark than the reference (hospital J). A similar pattern was observed in the multiple linear regression model. These differences could be targeted by policy interventions aimed at their reduction, taking into account the different case mix between hospitals of first and second level.

Interpretation of findings.
LoS is an easily available indicator of hospital activity, being an indirect estimator of resources consumption and efficiency. The hospital variability we found on Los post CS may be due to a number of factors, including differences in practice pattern, service efficiency, discharge policies, experience/ ability of obstetric staff and patient/family preferences 35 . Hospitals A and B are referral centres normally managing more complicated and serious obstetric conditions and some women delivering in the latter two centres may live quite far, hence these logistic barriers may push obstetricians to retain women admitted longer. By contrast, LoS was lowest for centres C and J, both located in the same local health unit (LHU) of FVG. The latter LHU provides domiciliary services to puerperae unable to go to hospital for a check-up during the first 10 days following ED for childbirth. These home visits are conducted by community midwives operating in health districts affiliated to the latter LHU.
Decreasing LoS inarguably increases demands on community postnatal services, the quantity and quality of which appears to vary globally 36 . For example, in Iceland, women are offered 8 home visits in the first 10 days postpartum, and their feed-back on postnatal care is generally positive 37 . By contrast, in Australia women are meant to receive at least two weeks postnatal support within their homes but continue to report low satisfaction with postnatal care as compared to antenatal and intrapartum services 38 . In the UK community postnatal care is provided by midwives, and although the National Institute for Health and Care Excellence (NICE) previously recommended a minimum of three home contacts post-childbirth 39 , many women are now asked to attend postnatal clinics instead, and there are no standards regarding the total number of post-partum contacts women should receive 40 . As such, wide variation is found in the number of postnatal contacts experienced by new mothers in the UK. A recent report from the UK National Maternity and Perinatal Audit (NMPA) project team found that the number of planned postnatal contacts for healthy women and babies ranged from 2 to 6, with a median of 3 41 . In an earlier survey of the Royal College of Midwives (RCM), 14% women in the UK reported that they only received one visit and a small minority reported no visit whatsoever 41 .
Interestingly, in the present study LoS > ED was less likely with increasing calendar year and with CS history, whereas it was far more likely among non-EU mothers. This suggests a positive approach of health care providers of FVG in decision making on LoS post CS, with socio-demographic and obstetric history factors probably taken into account.
Non-Italian women may have less family support, therefore may have benefited from longer LoS in FVG for a number of reasons, including inception and adaptation to breast-feeding. However, the impact of nationality and ethnicity may vary by type of health system. In countries adopting the voluntary health insurance (VHI), as the USA, the underlying dynamics on LoS may probably be different. For instance, findings from a secondary analysis of the Maternal-Fetal Medicine Units Cesarean Registry on 26,000 low-risk American women with singleton pregnancies, liveborn at 24-40 weeks, known ethnicity, up to 2 prior CS, and scheduled obstetric surgical procedures concluded possible disparities in quality and efficiency of obstetric care delivered to minorities 49 . Non-Hispanic Black women were more likely to incur longer LoS in the latter study, even after stratification by gestational age and type of CS, whereas Hispanic mothers had significantly shorter LoS across all gestational ages 42 . In another postnatal survey in 19 USA states during 2000, using data from the Pregnancy Risk Assessment Monitoring System, ED was more likely among Hispanic and Black women 43 . Lastly, in another populationbased postnatal survey conducted in 1999 on 2828 Californian women with low risk singleton pregnancies, ED was associated with lower socio-economic status, with untimely follow up more likely among latinas and non-English speaking women 44 .
Following inter-hospital variability, calendar year, number of previous livebirths and nationality of the woman, in the present study prolonged LoS after CS was influenced by child size factors. In particular we found LBW (2.0-2.5 kg) and pre-term gestations (33-36 weeks) both being strong determinants of prolonged LoS after PCS as well as UCS. A huge fraction of overall neonatal costs are reportedly leveraged by LBW and/or premature babies 45 , accounting for half newborn hospitalizations and 25% pediatrics costs in the USA 46 . In addition to decrease mortality/morbidity, interventions to delay or prevent premature deliveries could have a major impact on the containment of pediatric and newborn expenditures 46,47 . In a California study on 518,704 deliveries from the 2000 birth cohort, total adjusted hospital costs and LoS were calculated for both mothers and infants 45 . Total hospital costs for mothers comprised adjusted inpatient costs for any antenatal admissions as well as for postpartum hospitalizations, whereas for newborns they included adjusted inpatient costs associated with childbirth and with following hospital accesses (transfers or re-admissions) prior to primary discharge or before death, in case of child's decease before discharge. Whilst newborns weighing > 2500 g at birth had a mean LoS of 2.3 days, the respective estimate for LBW infants varied extensively from 6.2 to 68.1 days 45   www.nature.com/scientificreports/ for 5.9% births but 56.6% costs. We did not have information on hospital costs associated with childbirth, also because lack of information on sensitive data prevented the follow-up of infants across hospital registries. However, LBW accounted for 4.1% of all births and 8.6% OCS, whereas VLBW were 2.2% out of all deliveries and 7.2% of all OCS in the present study. Various pre-existing obstetric conditions as well as potentially preventable peri-surgical complications are associated with extended LoS post CS according to the open literature, including labour induction, labour augmentation (by oxytocin administration), ruptured membranes > 24 h, and epidural analgesia [48][49][50] .
Although we did not find any association with labour analgesia, there is evidence that the type of anesthetic technique employed is a strong predictor of extended LoS after CS, with longer hospitalization found with administration of epidural than spinal analgesia 12 . A study investigated 1,619 women undergoing CS during 2002-2005 at Aretaieio Hospital (Athens, Greece) in relation to the type of anesthesia administered. Although the impact of general anesthesia on LoS post CS decreased over the years in the latter study, neuraxial anaesthesia for CS was associated with shorter LoS than general anesthesia, and it was also influenced by the skill/ability of the surgeon 51 . A study at Ochsner clinic in New Orleans (Louisiana, USA) examined 840 consecutive parturients over a 1-year period. Prolonged LoS after CS was observed in 14.3% deliveries and was influenced by the type of anesthetic approach employed and the amount of intraoperative fluids administered during CS 12 . Among 57,812 women undergoing CS in USA between 1999 and 2002, within the network of the National Institute of Child Health and Human Development, independent obstetric risk factors for prolonged LoS included peri-surgical morbidities (general anesthesia, uterine atony, transfusion, hysterectomy, endometritis, ileus, wound and hemorrhage related complications), and perinatal conditions (pre-term gestation, birthweight). The most significant factors associated with extended LoS were ileus, endometritis and wound complications, but not general anesthesia 52 .
In the present study we cannot fully address the latter question, since until 2015 CEDAP data did not include details on the type of analgesia administered.
Hypertension/diabetes, pre-delivery LoS > 5 days and < 4 obstetric checks in pregnancy were equally associated with longer LoS post both PCS and UCS. Hypertension and eclampsia were factors significantly associated with longer LoS post CS also in the above study on 840 women undergoing CS at Ochsner clinic in New Orleans 12 . Pre-eclampsia and severe eclampsia (along with decreased gestational age, vaginal bleeding in the second half of pregnancy and suspected intrauterine growth retardation) are recognized prenatal factors associated with extended LoS [48][49][50] . The clinical conditions of the woman during pregnancy, including also pre-existing medical disorders (e.g. cardiovascular, respiratory, infectious, neurologic, autoimmune disease, etc.-factors not considered in our study) seemingly influence also the risk of readmission. For instance, in a USA study using the Healthcare Cost and Utilization Project's (HCUP) Nationwide Readmissions Database on 65,401 women affected by pre-eclampsia undergoing CS during 2014, 1016 (1.6%) had to be readmitted for hypertensive disorder and 90.6% of these readmissions occurred during the first 10 days following discharge. In the latter study longer LoS (> 5 days) was associated with lower adjusted risk of readmissions for hypertensive disorders within 60 days after discharge. Postpartum care is critical in determining the subsequent risk of readmission for sequelae related to eclampsia and pre-eclampsia, hence longer LoS following CS may be recommended in these conditions 53 . We could not fully confirm such findings related to readmission because of confidentiality of sensitive patients' data.
Although with minor level significance, LoS > ED was more likely in all days but Tuesday, with higher level of significance for Wednesday and Thursday. In Italy the civil registration offices are closed on Saturday and Sunday, therefore despite women delivering on Wednesday, Thursday or Friday may potentially be eligible to be discharged over the week-end, they are retained in hospital until Monday, when the will be able to register their child at the city council. By contrast, women delivering on Monday or Tuesday are more likely to be discharged by Friday.
CS performed during spring months (March-May) were associated with LoS > ED for both PCS, UCS and PCS, whereas the adjusted mean LoS during these 3 months was significantly higher only for PCS. Although with relatively weak significance, these findings slightly deviate from a previous study reporting higher risk of prolonged LoS post VD during winter (December-February) as well as spring months (March-May) 9 . The impact of cold weather and related morbidity would in fact be expected to be higher during winter months, where temperatures are usually lower and the risk of respiratory infections (especially influenza) higher. However, despite being lower than spring months, the crude rate of LoS > ED was still higher during winter months as compared to summer and autumn months for all three types of CS (OCS, PCS and UCS). Moreover, within spring months there was a declining trend of LoS > ED for OCS from March (46.8%), to April (44.7%) and May (43.6%). For  Hospital costs. Italy, which offers universal health coverage, is among the growing number of countries adopting a prospective payment system based upon capitation grants and diagnosis-related groups (DRGs), which fix the payments by estimated costs of hospital care ahead of service delivery. The DRG system has the advantage of stimulating the provider to contain the cost for each medical service, including unnecessary days of prolonged LoS 4 . The contingency capacity, bed turn-over and rationalization of available resources may have different impact on various hospitals. Nevertheless, in all multivariable models LoS was not influenced by number of admissions and number of births on delivery day at regional level, suggesting no impact of bed turnover on LoS. Governmental investments should be allocated to encourage measurements and controls of such differences, in order to maintain equity of health outcomes and costs across maternity services.
Prospects. The desirable model of obstetric care should be patient-centered and should deliver high quality of medical services yet containing health care costs by minimizing unnecessary prolonged LoS. Various models of postnatal management have been studied, delivering home-based, outpatient or inpatient services. These models consider and pursue different endpoints, including patient satisfaction, breastfeeding rates, health care costs and hospital readmissions for both women and newborns [54][55][56][57] . Integrated programs of primary and secondary care services, entailing frequent follow up home visits post hospital discharge (conducted by community midwives, nurses and/or general practitioners) seem capable of diminishing hospital re-admissions whilst ensuring quality of care and patient satisfaction. Nonetheless, these models of care may not be accessible and deliverable in every community setting, since they may be demanding in terms of organizational and human resources 57 . As a result, since it depends on the capacity to discharge patients into the surrounding non-acute facilities, LoS could become a debatable indicator of hospital 'efficiency' , as its variation could be explained by the characteristics of the hospital catchment area 58 . An interventional community-based outpatient postnatal clinic, the Monarch centre, was set up at Ottawa hospital (Canada) during 2014, with the aim to provide coordination between hospital care, community and primary care services. Pre-booked appointments were scheduled within 48 h of hospital discharge following childbirth. A number of services were provided, including mood screening/management, neonatal care, laboratory testing, breast-feeding assessment and support. General practitioners, lactation consultants and registered nurses were available for consultation on appointment. Out of 16,023 deliveries occurring between January 2012 to December 2016, the mean LoS was 46 h (66 h after CS vs. 37 h post VD). Eighteen months after the intervention, the average LoS for CS decreased by 20 h (significantly reducing by 27%); LoS post VD instead decreased much less (6 h), by 18%, but it was still significantly. Readmission rates of neonates at 30 days post discharge just rose from 1.1 to 1.9% 22 . Therefore, the implementation of integrated primary and secondary care services seems the key approach to contain unnecessary prolonged LoS after CS.

Strengths and limitations
Strengths of this study have been outlined elsewhere 8,9 .
Because for years 2005-2015 the CEDAP questionnaire collected time of childbirth but not hospital discharge's, we had to use day metrics instead of hours to estimate LoS. Although this is an important limitation, as differences in hours of LoS may have an impact on wellness of the woman and her family, the calculation of hospital costs by LoS in Italy is based upon days. However, in future it would be important for CEDAP to accurately record information on time of admission, time of birth and time of discharge.
As explained above, we did not have information on the address of residence of the woman, a logistic aspect that may have a major influence on decision making on LoS if the new mother lives far from the respective delivery facility. This is particularly the case for the two referral centres, which presumably receive more women from distant locations of FVG or even outside. It would therefore be very important in the future also to take into account the council of residence of the woman, in compliance with the Italian privacy law.
Although labour induction was limited to 15.6% out all deliveries in our studies, in the future it would be important for CEDAP to distinguish PCS from CS for failed induction. Moreover, UCS should be separated from CS during labour.
In the future it would be important for CEDAP also to collect information on other factors that may have an impact on LoS: type of analgesia administered during delivery; smoking status; body mass index (BMI); physical activity; amount of bleeding during delivery; confidence of the mother with breastfeeding and her readiness for discharge, Finally, although our database had a high level of completeness and accuracy of data, some important sociodemographic information (as father's education, father's occupation and marital status) were affected by a relevant number of missing values. Although this may reflect the woman's reluctance to reveal some personal (though anonymous) information, in the future it would be important to further improve the completeness of data collection by CEDAP, abandoning any form of paper document in favor of a standardized regional software for real time check-up of data entry, preventing input of inconsistent and/or conflicting data.

Conclusions
Variability of practice pattern by maternity centres confirmed to be the major driver of variability of LoS following childbirth in FVG.
Various organizational options are available to contain LoS after CS and reduce avoidable health care cost whilst maintaining and even improving the efficiency and quality of postnatal care. A planned contraction in the Scientific Reports | (2020) 10:19238 | https://doi.org/10.1038/s41598-020-74161-2 www.nature.com/scientificreports/ number of hospital beds, combined with the implementation of primary care services could contribute to effectively reduce the average LoS and apply policies of ED after CS, as successfully accomplished in some countries. Further in-depth interventions to achieve cost-effective obstetric outcomes could entail limiting the recourse to CS in absence of any clinical indication, changes in the hospital payment system and higher coordination of diagnostic and treatment paths within each maternity unit.