The study was carried out in Emilia Romagna, a region located in North-Eastern Italy with about 4.4 million inhabitants and approximately 40,000 births per year. The national health service is statutorily required to guarantee the uniform provision of comprehensive care throughout the country. Essential health services are provided free of charge, or at a minimal charge. Local Health Authorities, that roughly coincide with administrative units (provinces) are responsible for the overall health of, and for the services offered to the population.
The HDR of the women who delivered in the Emilia Romagna from 1st January 2007 to 30th June 2009 were linked to the BC using the HDR identification code, the hospital code and the year of delivery.
The HDR include demographics (ID number, sex, date and place of birth, place of residence), discharge ID, admission and discharge dates, discharge diagnoses and procedures (International Classification of Diseases, 9th Revision, Clinical Modification ICD-IX-CM), ward(s) of hospitalization, date(s) of in-hospital transfer, and the regional code of the admitting facility.
BC include demographic data of the mother, information on presentation and multiple pregnancy (singleton cephalic, singleton breech, transverse or oblique lie, etc.), parity (nulliparous, multiparous), the course of labour and delivery (spontaneous labour, induced labour or CS before labour) and gestational age (defined as the number of completed weeks at the time of birth) and other information on the newborn.
HDR were identified by using DRG codes (370–375), diagnosis (ICD-9-CM 640.xy - 676.xy (y = 1,2), V27 ) and procedure codes (ICD-9-CM 72.xy - 74.xy).
Records of women with a previous cesarean (ICD-9 CM code 654.2), discharged from hospitals without an operating room or small hospitals (<150 deliveries per year) or with a diagnosis of intrauterine death or still births (ICD-9 CM codes 656.4 V27.1, V27.4, V27.7) were excluded.
Primary CD were defined using procedure codes (ICD 9CM codes: 74.0, 74.1, 74.2, 74.4, 74.99 ) or diagnosis codes (ICD-9-CM 669.7) or DRG codes (370–371) or one BC variable (type of delivery).
We created 4 predictive models for primary CD (the Additional file 1
shows all variables of the four models and sources of information) using logistic regression:
Model 1 included age and maternal comorbidities and information about delivery recorded in the HDR of the index hospitalization
Model 2 included, in addition, past comorbidities recorded in HDR of hospitalizations occurred two years before delivery
Model 3 included additional clinical variables retrieved from the BC
Model 4 included also socio-demographic variables retrieved from the BC.
We applied a backward stepwise procedure to model 4 to identify the subset of variables significantly associated with caesarean section. Adjusted Odds ratios (OR) with 95% Confidence intervals (CI) were calculated for all models.
We evaluated the models at both patient and hospital level by comparing the full model (model 4) including variables derived from the two data sources with more parsimonious models, following the procedure of Dimick et al. [20
]. In particular, to evaluate patient level risk prediction we used:
the Receiver Operating Characteristic (ROC) to assess how well the model discriminates between women with and without a CD. The area under the curve ranges from 0.50 (no ability to discriminate) to 1 (perfect discrimination).
the AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion), to obtain measures that combine fit and complexity of the model. Lower values indicate a better fit of the model taking complexity into account.
To evaluate hospital level risk adjustment, we compared CD measures obtained using the 4 models. To this purpose, we calculated the ratio of observed to expected CD (“O/E ratio”) at each hospital. The O/E ratio was calculated using logistic regression to predict a probability of CD (i.e., the expected outcome) for each woman. These probabilities were then summed for every hospital. The observed number of events was divided by the expected number, to obtain a risk-adjusted estimate of the outcome of interest; an O/E ratio of 1.0 is as expected given that hospital’s women characteristics, less than 1.0 is better than expected and greater than 1.0 is worse than expected. We calculated O/E ratios using the 4 models to estimate the expected CD.
We then analysed the correlations between the hospital O/E ratios obtained with different models. A correlation coefficient of 1.0 implies perfect agreement in O/E ratios .