In this population based study involving in32, 280 cases, we found a small but significant increase in the incidence of RTHA from 2001 to 2008 in the Spanish population, from 20.2 to 21.1 procedures per 100, 000 inhabitants (p < 0.01). This represents a 3.75% overall increase, which is similar to other reports in developed countries .
In this period a total of 161, 791 discharges of patients having undergone primary total hip arthroplasty were identified and the overall crude incidence increased from 99 to 105 THA per 100, 000 inhabitants (p < 0.001) .
Kurtz and Mowat analyzed the National Hospital Discharge Survey (NHDS) of the USA, 1990 through 2002, to study the changes in the revision burden. The rate of RTHA increased by 3.7 procedures per 100, 000 persons over a 10 year period .
In the United States and Canada the revision burden, which refers to the percentage of revision hip replacements relative to the total number of primary and revision hip replacements, stayed roughly the same in the USA (14%-17% from 1993 to 2005) and in Canada (11%-13% from 2001 to 2006) , reflecting an increase in the absolute number of revisions as the number of primary procedures also increased.
This same trend is observed in some National Registries. In Norway the revisions constituted 12.3% of all the operations in 2003, 13.6% in 2007 and 14% in 2008 .
In 2009, the number of hip replacements reported to the Australian Registry increased by 1, 100 (3.4%) compared to 2008. Primary THA increased by 4.0% and RTHA by 1.1%. From 2003 to 2009 primary total hip replacement increased by 32.5% and revision hip replacement 9.3%. Remarkably, revision hip replacement as a proportion of all hip replacement procedures declined from 13.0% in 2003 to 11.2% in 2009 .
In Spain the ratio RTHA/THA has remained stable with figures around 20% from 2001 to 2008. The equivalent percentages for the Norwegian Arthroplasty Registry were 14.94% (922/6170) in 2001 and 16.37% (1114/6804) in 2008 .In Australia in 2004 the ratio was19.25% (3494/18153) decreasing to 16.63% (3677/22109) in 2008 
In our study the results of the Poisson regression analysis confirm that the increase in incidence of RTHA in men and also in women became greater after adjusting for potential confounders (age, sex and CCI). Although all groups increased in incidence those patients aged 75-84 or 65-74 experienced the highest (13.9 and 9.5 with 95% interval confidence of 13.3-14.5 and 9.1-9.9, respectively). Other studies using multivariate models have reached the same conclusions, verifying that the increase in incidence of RTHA is not only a consequence of population growth or ageing [3, 4, 22].
In a recent study Cram et al, in an observational cohort of 1, 453, 493 Medicare beneficiaries who underwent THA between 1991 and 2008 in the USA, observed that the mean age for patients increased from 74.1 to 75.1 years (P < 0.001). . In that same population and time period, among 348 596 subjects who underwent RTHA, the mean age also raised from 75.8 to 77.3 years (p < 0.001). .
In Spain, the number of high-risk surgical patients has increased over the last 8 years as shown by the analysis of the CCI. In 2001, 19% of patients had a Charlson Index of 1-2 or > 2. In 2008, the proportion of patients who had undergone a RTHA and had Charlson Index of 1-2 or > 2 had increased to 24.6% (p < 0.001). This same trend has been previously described in other studies and in the present series in the same period in THA [12, 19, 24].
A higher severity of illness store has been reported as predictive of a higher resource utilization for both primary and revision arthroplasty .
Our co-morbidity index figures for primary THA and RTHA are surprisingly similar. Cram et al  reported a significant mean increase in the number of comorbid illnesses per patient (from 1.0 to 2.0 for THA and 1.1 to 2.3 for RTHA) . The similar morbidity index observed in our patients undergoing primary and revision THA could be explained by survival (only healthier subjects survive long enough to need a revision) and selection (surgeons only conduct RTHA among those patients with low co-morbidities) bias. Further studies should be conducted to verify this.
LOS decreased from an average of 20.6 days in 2001 to 19.1 days in 2008 in Spain (p < 0.01), which is a small but relevant change.
In the same period, the LOS for THA significantly decreased from an average of 13 days in 2001 to 10.45 days in 2008 .
The mean nationwide LOS for Spain is longer than that described in other countries although a wide variability has been reported [23, 26].
As also suggested by other authors, we hypothesize that the reasons for observing a decrease in the Spanish LOS overtime may include: the presence of a larger rate of specialized departments using more efficient means to rehabilitate and discharge patients and an increased rate of discharges to short and long-term care facilities [26–28]. After adjusting by age and sex, our rate of patients that were discharge to health or social institutions significantly rose from 4% to 6.4% from 2001 to 2008.
The total costs of RTHA in Spain during our study period increased by 114.4%, from 34.8 million Euros to 74.6 million Euros. After adjusting for inflation, the average costs per patient increased by 78.3%, from 9, 375 to 16, 715 Euros. The total costs of primary THA In Spain during this period increased by 75%, from 120.6 million Euros to 211.34 million Euros .
Stargardt studied the variations in the cost of THA between and within nine member states of the European Union (EU), including Spain. The main cost drivers were found to be implants (34% of total cost on average) and ward costs (20.9% of total cost on average) .
More complicated revisions, like those requiring bone grafting, exchange of both components or specific complications, like periprosthetic fractures or infections, require higher resource utilization than easier ones [25, 30–33]. Even if we don't have data detailing the reason for revision we believe that the use of specialised 'revision-implants' is in part responsible of the large increase in the average cost per patient overtime in Spain.
The overall IHM after RTHA in Spain ranged from 1.16% in 2001 to 1.77% (p: 0.025) in 2008 (1.40% in women and 0.80% in men in 2001 and 2.02% in women and 1.44% in men in 2008).
Zhan and Kaczmarek screened the hospital discharge abstracts National Hospital Discharge Survey (NHDS) of five states of the EU during the year 2003. Their reported IHM rate was 0.84% and their rate of readmission, for any cause, within thirty days was 8.48%. Advanced age and co-morbid diseases were associated with worse outcomes .
In RTHA, prognostic factors related to higher mortality rates or complications may not be as clearly stated as in THA. Older age and high CCI may be more consistent but others like complexity of the revision, infected RTHA, poor preoperative functional status or female sex also seem to be important. These prognostic factors should help to optimize indications for THA and to reduce the already staggering, yet growing, burden of RTHA in developed countries, compared to THA, with greater LOS and higher cost [11, 22, 30, 35–38].
To the best of our knowledge this is the first study in Spain regarding changes in the incidence, demographic characteristics, co-morbidity profiles, and in-hospital outcomes of patients undergoing RTHA. In the absence of a National Registry for Arthroplasty the incidence, IHM rates and cost estimation reported in this study provide the best available information. The main strength of the current study lies in a large sample size and standardized methodology maintained over the study period.
Nevertheless, the present study presents some limitations. First, a potential source of bias comes from relying on administrative registries as several discrepancies between administrative data and audited and validated clinical data have been suggested [39, 40].
In Spain the CNBD was implemented in 1996 by the Ministry of Health in liaison with all of the autonomous communities. From that time efforts were made to improve the quality of the information including: periodic publications; mandatory educational programs for the persons responsible for the codification in the hospitals and periodical external quality control audits. Previous Spanish studies have assessed the validity of CMBD data using medical records as a reference, reporting that the CMBD is reliable for diagnosis and for estimating adjusted mortality rates. [41, 42].
Second, even if administrative data generally agrees with patient chart data for recording of comorbidities, it has been found that comorbidities tend to be under-reported in administrative data [43–45].
This could explain the large healthy population found in our study, based on our CCI. With regard to the use of the Charlson index to measure comorbidities, Burgos E et al conducted an investigation to assess the predictive value of six functional status and/or surgical risk scoring systems, including the anesthesiologic risk assessment instrument (ASA) and the Charlson index, with regard to serious complications after hip fracture surgery in the elderly. They found that all the scoring system reached a sufficient predictive value with regard to serious post-operative complications . However the similarities and differences between the Charlson comorbidity index and other commonly use score such as the ASA classification are discussed by Weaver et al. These authors found that among patient undergoing joint arthroplasties the discrepancy between the comorbidity index and the ASA was striking .
Lastly, outcomes were limited to the variables coded. The lack of differentiation between types of RTHA procedures in the current ICD-9-CM procedure coding system limits the utility of these codes in evaluating differences in patient and procedure characteristics in large public data. Other confounding relevant variables such as surgeon, hospital volume, cause leading to revision or the percentage of one-stage and two-stage revisions could not be analyzed as these variables are not collected in the database This lack of information about different brands thus unable to pick-up specific implant-related failures, compared with the National Joint Registries [20, 21].
Therefore, outcomes such as LOS and discharge destination may have been influenced by other covariates different from postoperative complications. In such a scenario, only IHM or LOS can be used to draw direct conclusions on the complication rate in the current study.
As noted above, because we used anonymised data it is impossible to detect double registrations, readmissions and transfers using the CMBD. Furthermore, unfortunately, in Spain we have no estimation of the number of double registrations, readmissions and transfers among patients undergoing a hip arthroplasty so a correction factor including this information could not be used to adjust the data. This is relevant because previous studies conducted in other countries have shown that the number of readmissions and transfers may have changed over time and may differ with age and sex [48, 49].
We believe that, notwithstanding its limitations the CMBD is a valid instrument to conduct epidemiological studies and has previously been used for this purpose, including by other authors. [20, 50, 51].
Although DRG have been a useful patient classification system for hospital cost analysis, DRG present a series of limitations [52, 53]. Riley in a recent review concludes that because administrative data have been collected for other purposes it is therefore not necessarily in a format that is intelligible or convenient to researchers. Furthermore, coding of diagnoses and procedures are more closely related to billing requirements than to medical records.