Data source
This study utilised administrative insurance claims data from the Statutory Health Insurance (SHI) sample of AOK Hessen (Versichertenstichprobe AOK Hessen/KV Hessen) [23]. Hessen is a state in central Germany that includes the major cities of Frankfurt and Wiesbaden. The population of the state was estimated at six million individuals in 2012; of these, 1.5 million were insured by AOK. The sample available for research (SHI) is acquired by drawing a random sample of individuals insured by the AOK with a constant selection set of 18.8%. The current SHI sample used in this study included 353,284 persons who were insured in AOK Hessen for at least one day during the five-year period of 2009–2013. The sample is population-based without disease-related selection, with no disease-related dropouts, no recall bias, and a high level of data reliability; this enables patient-based observation and a bottom-up approach to disease costing from the perspective of the health insurance fund. The SHI dataset contains details on healthcare transactions related to insured persons and healthcare providers, including data on care received in general practice, outpatient care (all specialist visits), and hospital care, including emergency visits. Details of this database have been previously published [23, 24].
Study population
We included mothers in the SHI sample who had a recorded diagnosis-related group (DRG) delivery code and a German procedure classification (Operationen- und Prozedurenschlüssel [OPS]) delivery code (Additional file 1) in the relevant study period (1 January 2009—31 December 2013). We further required women to be aged ≥ 12 and < 45 years at delivery and to have at least nine months of medical history available. We excluded women with more than one DRG delivery code within four months, those with no definite date of delivery, and those with a delivery discharge date in 2014. The index date was defined as the delivery date in the eligibility period. Women with multiple pregnancies during the study eligibility period were included in the study cohort once for each delivery, meaning that one woman could appear multiple times within the dataset. The baseline period was defined as the nine months preceding the index date. The delivery hospitalisation for mothers started from the day of hospital admission until the day of hospital discharge. Follow-up started from delivery hospitalisation discharge and lasted until the last date of data collection (31 December 2013), transfer out of the insurance fund, or the death of the mother, whichever occurred first. For women with multiple pregnancies, follow-up after each pregnancy lasted from hospital discharge until the beginning of the next pregnancy (defined as nine months [280 days] before the delivery date of the consequent pregnancy).
A cohort of PTL/PTB mothers was identified using any of the following criteria:
-
An International Classification of Diseases, 10th Revision (ICD-10) code indicating PTL during pregnancy and/or PTB (Additional file 2)
-
An estimated difference between the date of conception (calculated by the expected date of delivery) and the actual delivery date (defined by OPS codes) of less than 37 weeks
-
A DRG code indicating infant’s birthweight < 2500 g (Additional file 2)
-
For the subset of women who could be linked to their infant, delivery of a preterm infant using ICD-10 codes from infant records was also used to classify a mother as PTL/PTB.Footnote 1
It should be noted that not all mothers with PTL delivered a preterm infant, and delivery of a preterm infant was not a condition for being included in the study. This is because not all cases of PTL necessarily result in PTB [2].
Mothers who did not meet the criteria for PTL/PTB were considered non-PTL/PTB mothers. GA was defined using a recorded variable available within the database indicating the expected date of delivery. To calculate the GA, we used this expected date of delivery to calculate an estimated date of conception. This was done by assuming that all pregnancies’ estimated delivery date had been estimated as 280 days after the date of conception. By subtracting 280 days from the expected date of delivery, we derived an estimated date of conception. The difference between this estimated conception date and the actual delivery date was the GA. Additionally, ICD-10 codes present in the mother’s record during birth (P07.2 [extreme immaturity, GA < 28 weeks] and P07.3 [other preterm infants, GA 28–36 weeks] and O09 [duration of pregnancy]) were used to define GA. Mothers with missing GA were assigned to the > 37 weeks of GA based on the distribution of GA in the rest of the population.
Driven by the GA groups, as defined by ICD-10 codes, mothers were subsequently classified into three groups based on their infant's GA:
Study measures
Maternal characteristics
Data on demographics and clinical characteristics were assessed at delivery and during the nine-month baseline period. Characteristics of interest included age at delivery, plurality of births (multiple or singleton), infant GA, and maternal risk factors for PTL, [25] which could be captured in the AOK database through ICD-10 diagnosis codes: hypertension, diabetes mellitus, gestational diabetes, and depression. Information on baseline clinical conditions was used to calculate the updated Charlson Comorbidity Index (CCI) to estimate the overall health status of the mothers [26]. We used diagnosis codes recorded in the inpatient and outpatient setting to define the presence of clinical conditions and maternal risk factors for PTL.
Resource use and costs
Resource use and total direct medical costs (in Euros) were examined during pregnancy, at delivery hospitalisation, and up to three years post-delivery. During each period, outpatient resource use, outpatient prescription data, inpatient resource use, and other services (defined below), which included all services not reimbursed in the inpatient or outpatient setting, were considered. Specifically, we considered the following resources in the outpatient setting: laboratory tests, preventative procedures (such as cancer screening and vaccinations), basic procedures (such as ultrasounds, magnetic resonance imaging [MRI], or echocardiograms [ECG]), prescribed medications, general practitioner [GP] visits, gynaecologist/paediatrician visits, and any other specialist visit. It should be noted that due to the nature of the reimbursement system in Germany, which reimburses outpatient physicians on a quarterly basis, visits to physicians are recorded as one per quarter irrespective of how many encounters took place within the same quarter. Other services considered were: remedies (such as massages or occupational therapy), medical devices, midwifery services, driving services, and other (such as household help or home care). Outpatient costs were estimated as total costs for each service used or drug prescribed. In the inpatient setting we considered the total number of all-cause hospitalisations, length of stay (LOS), pregnancy/labour procedures, diagnostic tests, or therapeutic procedures (such as operations) performed during hospitalisation. Inpatient costs were estimated based on DRG codes per hospital stay. Costs were estimated from the third-party payer perspective, corresponding to the SHI fund, which is in accordance with Institute for Quality and Efficiency in Health Care (IQWiG) guidelines for cost of illness analyses in the German setting [27].
As not all mothers were insured for 365 days in the respective years of follow-up (lost to follow-up or reached the end of the observation period), costs for the first, second, and third year of follow-up were evaluated for those mothers who had sufficient follow-up and were continually insured in the respective year. It should be noted that we allowed women to enter the cohort up until the end of the study period (31 December 2013)—this means that only women enrolled prior to the 1 January 2011 could accrue the full three years of follow-up, and among these, only those who did not die or were lost-to-follow-up (LTFU) were observed for the full three years. We nonetheless chose to include women who could not be followed for the full three years to maximise the study cohort available for analysis.
Statistical methods
Continuous variables were described using average values (median and mean) and measures of data dispersion (interquartile range [IQR], minimum and maximum values). Categorical variables were described using frequencies and percentages. P-values comparing the characteristics of PTL/PTB mothers to non-PTL/PTB mothers were calculated using the chi-squared test/Fisher’s Exact test in those instances when expected cell counts < 5. Univariable negative binomial models were used to estimate the rate of resource utilisation during follow-up as the number of events per person-year with 95% confidence intervals (CI). Costs were presented using summary statistics, which were estimated and included mothers without any resource use (i.e., median and mean costs were estimated, including women who incurred zero costs).
All data programming and analyses were carried out using Microsoft SQL server 2008 (Microsoft, Redmond, WA, USA) and SAS for Windows Release 9.3 (SAS Institute Inc., Cary, NC, USA).