Study design and setting
This matched cohort study utilised data from the National Joint Registry for England, Wales, Northern Ireland and the Isle of Man (NJR), which was linked to inpatient and day case admission data from the Hospital Episode Statistics (HES) between April 1, 2003, and December 1, 2014. HES data includes data on inpatient and day case admissions in England funded by the English NHS, as such, the analysis was limited to patients receiving NHS-funded treatment in England.
In this study, we aimed to compare the inpatient and day case costs of patients who underwent one- or two-stage revision THR for PJI following their primary THR (revised PJI patients hereinafter) compared to matched patients whose THR was either not revised or revised for reasons not related to PJI (comparator patients hereinafter).
Patients were eligible for inclusion in the revised PJI group if one of the indications for revision was recorded as infection by clinicians in the NJR at the time of revision (more than one indication can be selected); they received a one-stage revision or at least part one of a two-stage revision for PJI between 2006 and 2009; the surgery was the first revision for PJI on the index side (the index side refers to the hip side that is included in the analysis, for revised PJI patients it is the side with the hip that was treated for infection); their primary THR could be identified in the NJR; they did not have revision surgery for PJI on the non-index side during the 5 years following THR surgery for their index side; the revision surgery for PJI was within 5 years of their primary; they had complete matching variables and their NJR records could be linked to HES.
Patients were eligible for inclusion in the comparator group if they had a primary THR between the dates of the primary THRs of revised PJI patients; they did not have revision for PJI on their index side (for comparator patients, the index side is the side with the hip that had a primary THR during the period of revised PJI patients THRs) reported in the NJR data (available until 2009); they did not have revision surgery for PJI on the non-index side during the 5 years following THR surgery for their index side; they had complete matching variables and their NJR records could be linked to HES. Comparator patients could have had revision surgery for indications other than PJI. Comparator patients may also have developed a PJI and received alternative treatments such as antibiotic suppression.
Once eligible patients were identified they were matched using a combination of exact and radius matching with a matching ratio of 1 revised PJI patient to 5 comparator patients, without replacement. To maximise the sample size, where less than 5 comparator patients were identified, the revised PJI patient and matching comparator patient(s) were still included. Matching variables were selected from patient characteristics and primary THR surgery factors that previous research suggests potentially impact the likelihood of PJI following THR [22, 23]. The matching process incorporated exact matching for some variables (sex, ASA grade, type of hip replacement (total or resurfacing), hospital) and radius (close) matching for variables where we were unlikely to find exact matches (date of primary THR, age). We allowed a radius of plus or minus 1 year for the date of primary THR and plus or minus 10 years for age.
Identification of resource-use and estimation of cost
For all patients, inpatient and day case admissions (not limited to orthopaedics admissions) reported in HES for the 5 years following their primary THR was cleaned and processed through the HRG4+ 2014/2015 Reference Costs Grouper  to obtain Healthcare Resource Groups (HRG’s). An HRG is a group of clinically alike treatments that use similar levels of healthcare resource. Cleaning included identifying and removing duplicate records and reformatting HES records to enable them to be processed by the Grouper. NHS reference costs were used to estimate costs; reference costs are based on the average unit costs of NHS providers . HRGs are costed by applying reference costs to the core HRG, unbundled HRGs and excess bed days. Spell-level reference costs, where spell refers to a single hospital stay from admission to discharge, were applied to each spell HRG using NHS reference costs . Where the Grouper did not provide an HRG, weighted average costs of adult HRGs by admission type (elective, non-elective short/long stay, regular day/night, day case) were applied. Costs of each HRG spell were then summed to estimate 5-year costs following primary THR.
All statistical analyses were performed in Stata 15.1 (StataCorp LLC, College Station, TX ). Multiple model specifications were explored and compared using, but not limited to, histograms, quantile-quantile and percentile plots of deviance residuals and Akaike’s information criterion. A two-part model [27, 28], which accounted for clustering of revised PJI and comparator patients within their matching group, was employed using the twopm Stata command  to estimate the difference in number of stays and costs. To account for excess zeros, in the first part of the model, a Probit model was used to estimate the probability of total costs equalling zero. In the second part, a generalised linear model was used to assess the distribution of costs in revised PJI and comparator patients who had at least one inpatient or day case admission, with an identity link function and gamma distribution to account for the positively skewed distribution of costs. Age, sex, ASA grade, diagnosis of osteoarthritis, operation date, Charlson Comorbidity Index, bearing surface (metal-on-polyethylene, metal-on-metal, ceramic-on-polyethylene, ceramic-on-ceramic, metal-on-ceramic/ceramic-on-metal) and procedure (cemented, uncemented, hybrid, reverse-hybrid, resurfacing) were controlled for within the model. To account for the intragroup correlation of patients within matched groups, a variable indicating the matched group was included as a variance estimator cluster option in the model. Further information on the model specification is provided in Additional File 1.