Incidence and costs of cardiac device infections: retrospective analysis using German health claims data

The analysis was based on anonymized billing data collected from German statutory health insurances (SHIs). The database used for this analysis comprised 2.8 million people, 3.5% of the nationwide population. It was adjusted with regard to age, sex and morbidity distribution so that it was representative of the German SHI population. In Germany, health claims are submitted by healthcare providers working for the SHI. These billing data include the ICD-10-GM (German modification of International Classification of Diseases) codes for the classification of diseases, the German version of the International Classification of Procedures in Medicine codes (OPS), the German diagnosis related group codes for inpatient billing, as well as the German national drug codes (ATC codes) for outpatient drugs prescribed. The data on costs were the amounts reimbursed by SHIs to healthcare providers (known in the USA as payment), thus using payment data gives the cost analysis the perspective of the SHI. Using these data is in accordance with the methods for health economic evaluations in Germany [13]. Average costs are presented for six different categories: inpatient care (hospitalization), outpatient care, prescribed pharmaceuticals (outpatient setting), sick pay (payment of a percentage of a regular salary when someone is unable to work), remedies and medical aids (such as massages, shoe inserts, walkers) and total costs. Inpatient care costs are based on German diagnosis related groups, which provide lump sum reimbursement payments to hospitals for whole inpatient episodes of care.

The objectives were to estimate the incidence of CDI in patients implanted with ICD or CRT-D devices and report associated healthcare payer costs. Patients eligible for inclusion had an ICD or CRT-D implant in the 4-year period 2010–2013 and all were followed up to 31 December 2014. At least 12 months’ continuous enrollment was required before the implant index visit.

A before-and-after analysis was completed for patients with CDI to characterize resource use and costs 12 months before and after CDI, with subjects acting as their own controls. The date of CDI served as the index date for this analysis. Second, a case-controlled analysis was performed to estimate incremental costs for patients with and without a CDI, using a propensity score matching (PSM) methodology. The date of CIED implant served as the index date for this analysis. Data were extracted for 12 months prior to implant with minimum and maximum follow-up postimplant of 12–36 months. Information on patient baseline characteristics, comorbidities, severity of infection and treatment pathways were analyzed.

Eligible patients had a record of an initial ICD/CRT-D implant or replacement (full system implant or generator only) during the calendar years 2010–2013 (list of implant procedure codes in Supplementary Table 1). CIED patients were classified in two mutually exclusive groups according to German OPS procedure codes: de novo implant or replacement implant.

For patients undergoing de novo implant, infections were counted if they occurred on the same day (outpatient) or during the same hospital stay of the device implantation or within 365 days thereafter.

For the replacement cohort, it was not possible to discern from hospital episode codes regarding a single hospital stay whether a patient underwent a replacement procedure in response to a CIED infection or if their infection arose within the same stay as a device replacement procedure. It was assumed that if an infection code was present for an index device replacement stay, the replacement was the result of an infection that initiated before the replacement visit. Hence for the device replacement cohort, infections were only counted if they occurred after the index device replacement stay.

CDI was defined by two conditions occurring ≤365 days after initial or replacement implant:

Once identified as a patient with a CDI, patients were retrospectively classified as having either a minor or major infection as follows:

The 8-week threshold was selected based on clinical practice.

A second method to discriminate severity of infection was to classify CDI patients into those treated in an ambulatory setting only (no hospital admission related to infection) versus those with a hospital admission.

Three types of statistical tests were employed. T-tests were used for calculating differences in means. Fisher's exact test was used for analysis of contingency tables, such as share of male patients in the CDI group compared with the group without CDI. Binomial tests were used for dichotomous data: for instance, 95% CIs for major and minor infections were calculated using a binomial test.

Mortality or switching of health insurance provider can reduce the size of patient cohorts over the course of 3-year follow-up in an SHI healthcare claims database. We calculated mean per patient costs based on cumulative costs per cohort divided by the starting size of each cohort.

Data were stored and analyzed using Microsoft Office Excel^® 2010 (Microsoft Corporation, WA, USA) and SAS^® (Version 9.2; SAS Institute Inc., NC, USA).

In order to estimate incremental costs associated with a CDI, patients with an infection were matched to similar patients who did not experience an infection using PSM techniques, which balanced potential confounding variables between the two groups [14]. The matching technique used was a kernel matching with an Epanechnikov function [15]. Kernel matching is a nonparametric matching estimator that uses weighted averages of all individuals in the control group (CG) to construct the counterfactual outcome. Using a weighted average instead of matching individual patients generally results in a lower variance. Weights depend on the distance between each individual from the CG [16]. Paired patients have their index event in the same year and the same duration of follow-up. The defined covariates for the matching have been set as follows: age, sex, de novo or replacement implantation, ambulatory or in-hospital implantation and the following costs measured over the 12-month period prior to device implant: mean outpatient costs, inpatient costs and sick pay.

Patients who had a record of an ICD or CRT-D implant or generator replacement during the period 2009–2013 numbered 4699 (3233 de novo and 1466 replacement). There were 158 CDIs in the 12 months after implantation, an annual risk of 3.4% (95% CI: 2.8–3.9%). The risk differed for de novo and replacement implants: 2.9% (95% CI: 2.2–3.3%) for de novo and 4.4% (95% CI: 3.6–5.8%) for replacements (p < 0.01), see Table 1).

Split by definitions of severity, 57.0% (95% CI: 49.2–65.3%) were minor infections and 43.0% (95% CI: 34.7–50.8%) were major infections. CDIs were treated with ambulatory care only in 40% (95% CI: 32.8–48.8%) of cases and in a hospital inpatient setting in 60% (95% CI: 51.2–67.2%) of cases. Considering infection severity and treatment setting in combination, 70% of all minor infections were treated in the ambulatory sector and 30% were treated in the inpatient sector; whereas all major infections were treated in hospital.

Patients who experienced a CDI, compared with those who did not (prior to PSM), had significantly higher rates of several known risk factors for CDI: kidney disease, having malignancies or being immunocompromised or taking oral anticoagulant drugs (Supplementary Table 2).

With patients acting as their own controls, costs were €15,822 higher per patient in the 12 months after infection (all severities) than the 12 months before after exclusion of the cost of index device implantation: 29,195 versus €13,373 (p < 0.05; Figure 1). The higher cost was driven by hospital inpatient care (98%).

PSM created infection groups (IGs) and CGs of equal size (146 patients each); 12 patients with CDI were unable to be matched to controls. Characteristics of the two matched groups were well balanced (Supplementary Table 2).

Estimated rates of healthcare utilization were greater in the 12 months following implantation for those with an infection compared with matched controls (Figure 2):

Estimated 3 year incremental healthcare expenditures (from date of implantation) for patients with an infection (IG) of any severity, compared with the CG patients without an infection, were €31,493 (p < 0.01) for de novo implant patients and €33,777 (p < 0.01) for replacement patients (Figure 3). The index date for this analysis was the implantation rather than the infection, thus if a patient had a CDI identified 11 months postimplant (for example), the costs associated with management of the infection spill-over into year 2; 81–84% of the difference in 3-year cumulative expenditure occurred within the 2-year period.

Cumulative costs per patient were higher for patients with a major infection (€99,706; CI: €81,632–117,780), than for patients with a minor infection (€52,540; 95% CI: €40,706–64,373) (p < 0.01). Incremental expenditures by infection severity compared with controls were €12,249 for minor infections and €59,419 for major infections.

The analysis was conducted using data from several SHIs comprising approximately 2.8 million people, 3.5% of the nationwide population. Similarly, the number of implants included in this study is estimated to represent 3.6% (4699) of the total number of ICDs implanted in Germany in the period 2009–2013 (129,000).

The analysis is based on anonymized billing data from a German health claims database; such databases were designed primarily to enable appropriate hospital financing and not for epidemiological purposes. The accuracy of the information depends on physician coding practices and the quality of data could differ between health units or regions.

Information was not available about periprocedural use of intravenous antibiotic prophylaxis in each patient, which significantly reduces CDI and is recommended practice [5–7,22]. Furthermore, microbiologic culture results were unknown for each case of CDI and therefore we were unable to demonstrate the role of different bacteria.

For the device replacement cohort, infections were only counted if they occurred after the index device replacement stay; for hospital stays where a CDI was documented with no prior record of a CDI in that patient, and a replacement procedure was conducted during the same stay, it was not possible to discern whether the patient came directly into hospital for a replacement procedure due to CDI, or the patient had come for a planned replacement (for instance due to battery depletion) and developed a CDI after surgery and prior to discharge. This limitation may underestimate the CDI rate for replacement procedures, and conversely overestimate the rate for de novo implants. However, this may only have had a small effect, because it is uncommon for CDI to be diagnosed within the same stay as the implant procedure [11,23].

We defined minor CDI as without a device procedure within 8 weeks after documented infection. However, 30% of minor CDI in this study were managed with inpatient care and we acknowledge a limitation of our definition and recognize that some minor CDIs were serious.

Mean cumulative costs per patient over 3 years were estimated without adjustment for a decreasing number of patients still present in the SHI database over the course of the follow-up period, due to mortality or switching of health insurance provider. This approach makes the results applicable for SHI companies to quantify the budget impact of CDIs. However, the average cost per patient based on the whole starting cohort, from an SHI perspective, underestimates the mean cost of surviving patients, which should be kept in mind for using these cost data in Markov-type cost–effectiveness models.