Total hip replacement is a paradigmatic quality-of-life intervention. It is not intended to increase the quantity of life (survival), but it does dramatically improve the quality of life. Nonetheless, Scandinavian researchers have found that the long-term survival of patients who have had primary total hip replacement is better than that for the general population1-6, and they have speculated that the reason may be patient selection for the total hip replacement, the use of anti-inflammatory drugs among patients with arthritis, a more active lifestyle after surgery, or some other factor. A few studies from the United States have described similar findings7-9, but those studies had short-term follow-up7,8 or the patients were from a single surgical center9. Few of the studies had an internal control group, and none that we are aware of tried to take into account the comorbidity status of both the patients and the controls.
In this report, we compared the six-year survival rate in a cohort of Medicare recipients who had total hip replacement in 1996 with that of a matched cohort of enrollees from the general Medicare population. We hypothesized that patients undergoing total hip replacement would have less comorbidity than controls and would also have better long-term mortality. We aimed to assess whether the association between total hip replacement and survival had the characteristics of a causal relationship or whether it might appear to be due to comorbidity or other factors.
Materials and Methods
Total Hip Replacement Cohort
We defined the total hip replacement cohort by using Medicare claims submitted by United States hospitals and surgeons for procedures performed in the six months from January 1 to June 30, 1996. A hospital claim with an ICD-9-CM (International Classification of Diseases, Ninth Revision, Clinical Modification) procedure code of 81.51 (total hip replacement) was required, together with a surgeon's claim in the same admission with a CPT (Common Procedural Terminology) code of 27130 (arthroplasty, acetabular and proximal femoral prosthetic replacement). Patients undergoing total hip replacement for fracture of the hip or other femoral fracture were omitted, as we wished to focus on elective procedures. These femoral fractures were identified by a hospital ICD-9-CM discharge diagnosis code 820-821.11; a hospital ICD-9-CM procedure code 79.05, 79.15, 79.25, or 79.35; or a surgeon's claim with a CPT code 27230-27248 or 27500-27507. We excluded patients enrolled in health maintenance organizations, those not enrolled in both Part A and Part B of Medicare, and patients who were not residents of the United States, as claims for these patients are often incomplete. We also excluded beneficiaries who were more than ninety-nine years old, since very few centenarian enrollees receive total hip replacement. To ensure that we had at least a year of Medicare claims before the total hip replacement, we omitted patients who were in the first year of Medicare eligibility (that is, the age of sixty-five). There were 28,469 primary total hip replacements performed in patients between the ages of sixty-six and ninety-nine years in the final cohort.
To select a comparison cohort of patients from the general United States Medicare population, we randomly allocated to all enrollees in the standard 5% Medicare sample the procedure date of one of the total hip replacement cohort members. If the population member was alive on the day before his or her allocated index date and was between sixty-six and ninety-nine years old, a resident of the United States, enrolled in both parts of Medicare, and not enrolled in a health maintenance organization, then he or she was retained. This procedure identified 1.28 million Medicare enrollees as potential controls. From this pool, the control cohort was randomly chosen and frequency-matched to the patients who had a total hip replacement in a 5:1 ratio by sex, year of birth, race (black, white, or other and/or unknown), and whether the state Medicaid program paid the Medicare premiums for the patient (a marker of poverty). The comparison cohort thus comprised 142,345 matched controls for the 28,469 patients in the total hip replacement cohort (Table I).
The annual Medicare denominator files contain the date of death of all enrollees who died during the year in question. All total hip replacement patients and controls were followed with use of these denominator files from the total hip replacement (or control index) date until death or December 31, 2001. Thus, all surviving enrollees were followed for a minimum of five and a half years and a maximum of six years.
We examined the discharge diagnoses from the Medicare hospital claims files for the 365 days before the date of the total hip replacement or the control index date and ascertained for each subject whether each of sixteen serious hospital diagnoses had been made. These diagnoses are listed in the Appendix, and they are the constituent elements of the Charlson comorbidity index10,11. This index is a single number formed by a weighted sum of the diagnoses, with the set of weights calculated according to their ability to form a single summary index that predicts death within a year in the Charlson datasets.
The numbers and percentages of patients managed with a total hip replacement and of controls who had had each of the hospital diagnoses were counted, and odds ratios with 95% confidence intervals were constructed. We used actuarial life-table methods to compare the mortality experience of the patients managed with total hip replacement and the controls over time. Two functions of survival were calculated. The familiar “survival curve” plots the cumulative probability of survival as a function of an increasing duration of follow-up. By definition, this probability cannot increase. The monthly hazard function, on the other hand, is an estimate of the death rate within a specified month of follow-up, for those who have survived to the beginning of that month. Unlike the cumulative survival curve, this death rate may fluctuate up and down over time. The overall survival experience of the two groups was compared with use of the log-rank test. The actuarial life-table technique, survival functions, and log-rank test are discussed in standard statistical texts12,13.
A proportional hazards model was created for each of the sixteen comorbid diagnoses, to assess the separate effect of each diagnosis on mortality. Each of these models contained the variables total hip replacement versus control, age, sex, race (black or white subjects only), and Medicaid status as well as the diagnosis in question.
We then constructed two proportional hazards models to examine whether total hip replacement had an independent effect on time to death. The first analysis included adjustment for sex, race (black or white subjects only), age, and Medicaid eligibility. In the second model, the Charlson comorbidity index was added. In both models, we included an interaction between the cohort type (total hip replacement or control) and the time after the total hip replacement date (or control index date). We estimated the relative risk of death for the two groups within three time-intervals: less than three months after surgery, three months to five years after surgery, and more than five years after surgery. Such interaction is a time-dependent covariate13,14. All analyses were carried out with use of the SAS statistical software, release 8.2 (SAS Institute, Cary, North Carolina).
Table I summarizes the demographic characteristics of the total hip replacement and comparison cohorts. Nearly two-thirds of each cohort (18,495 total hip replacement patients and 92,475 control subjects) were women, 95% (27,018 total hip replacement patients and 135,090 controls) were white, and 6% (1683 total hip replacement patients and 8415 controls) had their premiums paid by state Medicaid programs. The mean age (and standard deviation) at the time of the total hip replacement or at the index date was 74.8 ± 5.8 years for both cohorts.
As shown in the Appendix, virtually identical percentages—3490 (12.3%) of the total hip replacement patients compared with 17,944 (12.6%) of the controls—had had a hospitalization in the year before the index date with one or more of the comorbid discharge diagnoses listed in the Charlson index. Nevertheless, there was clear evidence that the control beneficiaries had worse health than the total hip replacement recipients. The controls were much more likely than the patients managed with total hip replacement to have multiple diagnoses; they were substantially more likely to have twelve of the sixteen diagnoses. Only peptic ulcers (odds ratio, 1.69; 95% confidence interval, 1.49 to 1.91) and rheumatologic disease (odds ratio, 2.67; 95% confidence interval, 2.34 to 3.04) were more common in the total hip replacement cohort.
Overall Survival of the Total Hip Replacement and Comparison Cohorts
Figure 1 shows the survival curves for the patients who had primary total hip replacement and the control subjects. Immediately after surgery, the total hip replacement cohort experienced greater mortality than did the controls; however, by ninety days, 309 patients who had a total hip replacement and 1603 controls (1% of each cohort) had died, and the survival curves crossed. The total hip replacement cohort had considerably better survival than did the controls thereafter. By five years of follow-up, another 32,022 (22%) of the controls had died compared with 4547 (16%) of the patients who had total hip replacement. In the sixth year after surgery, just over 3% (924 patients who had total hip replacement and 4938 control subjects) in each group died, showing that very little of the advantage of the total hip replacement patients remained. Altogether, after six years, the probability of survival was 79% for the total hip replacement cohort compared with 72% for the controls (log-rank, p < 0.0001).
Figure 2, the hazard function, shows the changes in mortality rates over time. It plots the monthly probability of death among the subjects in each cohort who have survived to the beginning of that month and is an estimate of the death rate within that particular month. For the total hip replacement cohort, the probability of dying in the first month after surgery was much higher than it was for the comparison cohort, but it fell abruptly and markedly thereafter. Within three months after surgery, it had dropped to well below the level of the controls. In subsequent months, the mortality of the total hip replacement cohort increased fairly steadily and approached that of the controls after approximately five years. For the comparison cohort, the probability of death showed monthly fluctuations around a pattern of increasing mortality with time, an expected pattern for deaths in an aging population.
Demographic factors had strong effects on survival (see Appendix). Overall, female subjects were much less likely to die than were male subjects (hazard ratio, 0.63; 95% confidence interval, 0.62 to 0.64). Subjects who were Medicaid-eligible had a considerably poorer survival rate than did those with higher incomes (hazard ratio, 1.87; 95% confidence interval, 1.80 to 1.93). Elderly subjects were much more likely to die than were younger Medicare recipients. As expected, each of the comorbid diagnoses also had a strong effect on mortality. Earlier hospital diagnoses of metastatic cancer, renal failure, liver disease, and leukemia and/or lymphoma in particular carried hazard ratios of 4.
Table II shows the mortality ratios for the total hip replacement patients relative to the controls, calculated with and without adjustment for comorbidity. After adjustment for age, sex, race, and Medicaid eligibility, the total hip replacement cohort had a better survival rate than the controls. For the initial three-month period, the risks of death for the two groups were comparable (hazard ratio, 0.95; 95% confidence interval, 0.84 to 1.07), suggesting that by ninety days after total hip replacement, the early deaths related to surgery were already counterbalanced by the favorable survival rate of the total hip replacement group. For the period of ninety days to five years after surgery, the hazard ratio was 0.65 (95% confidence interval, 0.63 to 0.67), indicating much better survival for the total hip replacement cohort. In the final year of follow-up, the hazards had begun converging (hazard ratio, 0.82; 95% confidence interval, 0.76 to 0.87).
Even though the comorbid diagnoses were important predictors of mortality, adding the Charlson comorbidity index to the model did not explain the survival difference between the two cohorts (Table II). In the first three months following surgery, the mortality risks for the total hip replacement patients and the controls were still comparable (hazard ratio, 0.97; 95% confidence interval, 0.86 to 1.10). The estimated hazard ratio for the period of three months to five years after surgery remained at 0.65 after comorbidity adjustment, and the estimated hazard ratio in the last year was 0.81 with comorbidity adjustment compared with 0.82 without adjustment. Similarly, adding all of the component diagnoses of the Charlson index as separate variables, instead of the index itself, had little effect on the hazard ratio (data not shown).
We compared the survival of a cohort of 28,469 United States Medicare beneficiaries who had elective total hip replacement with 142,345 controls from the general United States Medicare population. We found that the total hip replacement cohort had less major comorbidity than did the matched controls, with an approximately 30% lower prevalence for most serious diseases, as reflected in hospital discharge diagnoses in the year before surgery.
As might be expected after any major surgical procedure, the total hip replacement patients had a higher mortality rate than the typical Medicare patients in the first month after the procedure. However, the mortality rate in the total hip replacement group then dropped precipitously, and, by ninety days, the cumulative survival in the two groups was comparable. Subsequently, the mortality rate in the total hip replacement group was lower than that in the control group, although the mortality rates appeared to be converging after about five and a half years.
The low mortality rate among total hip replacement patients persisted in our analysis even after adjustment for sex, age, race, Medicaid eligibility, and major comorbid diagnoses as captured on hospital discharge data. It is possible that this survival advantage could be due to the surgery. After the procedure, the use of nonsteroidal anti-inflammatory drugs and narcotic pain medications typically decreases and patients are able to exercise again—factors that could improve their health. Also, the fact that they were once again in control of their own lives could conceivably affect their life expectancy15.
The mortality differential could also be due to unmeasured differences between the total hip replacement and control populations rather than to the effects of the procedure itself. We were able to use only hospital discharge diagnoses recorded on Medicare claims for the year before the index date. Only about 20% of each cohort (5966 total hip replacement patients and 25,132 controls) had a hospitalization in this one-year period, and it is likely that some of the major medical problems were missed. We also had no information on smoking status, body mass index, and other factors that have strong effects on mortality. Access to the patients' medical records, to earlier hospital claims, or even to patient self-reports would have increased our knowledge of the comorbid conditions16-18.
There was some evidence suggesting that the mortality differential we observed was due to patient-related factors rather than to the effects of the surgery. The real effects of a procedure such as total hip replacement are likely to require some time to become evident, whereas we observed a startlingly rapid improvement, with mortality rates falling substantially below those of the matched Medicare controls within three months (Fig. 2). A reasonable interpretation of this pattern is that, after the (small) perioperative mortality rate had run its course, the observed low death rates simply reflected the inherently low mortality rate of the patients selected for surgery. These differences may be reinforced by the preoperative evaluation process, as the medical history, physical examination, laboratory tests, and electrocardiogram serve as a screen to uncover potentially life-threatening conditions in time to treat them or to suggest that total hip replacement is not appropriate. Thus, patients who go on to have a total hip replacement have passed such a selection screen.
Our observations confirm and extend the observations made in the Scandinavian studies and in the earlier American studies that patients who have had a primary total hip replacement have a better survival rate than a matched population after the perioperative period1-6,8,9. One previous study also compared mortality in total hip replacement patients and controls in different time-periods after surgery and found, as we did, that the differences in mortality rate lessened over time. Among 24,638 Finnish total hip replacement recipients, the mortality rate was 44% lower than that of the general population for up to five years after total hip replacement and was 16% lower after five years or more1. After six years, approximately 20% of our total hip replacement patients had died, the same percentage as reported for patients who were sixty-five years old or more who had elective total hip replacement in Sweden4.
The major limitation of our study is the use of administrative data. Medicare inpatient claims data are sparse on clinical details, and comorbidity is a key potential confounding factor in this study. Our hospital-claims-based diagnoses were only a crude measure of major comorbidity, in part because only about 20% of the patients (5966 total hip replacement patients and 25,132 control subjects) had any hospitalization in the year prior to surgery. Further research should be done with use of data that permit more detailed comorbidity assessment.
The strengths of the study include the large sample size and virtually complete follow-up for at least five and a half years for both cohorts. Our control cohort was taken from the Medicare population rather than from published life tables, and so we were able to match simultaneously on sex, year of birth, race, and also on Medicaid status. Enrollees whose state Medicaid program pays the Medicare premiums are a third less likely to receive a total hip replacement, and nearly twice as likely to die in the first ninety days after total hip replacement, than are other enrollees19, so we controlled for this important confounder by matching. We were also able to use a series of comorbidity measures, however imperfect, for both patients and controls.
We concluded that older persons who have elective total hip replacement have a substantially higher six-year survival rate compared with controls matched for age, sex, race, and income status. We found consistent differences in major comorbidity between total hip replacement recipients and control, but the improved survival rate for the total hip replacement cohort persisted despite our admittedly crude adjustment for measured comorbidity. The survival differences emerged so rapidly after the procedure that they seem to reflect inherent characteristics of the patients, such as unmeasured comorbidity or the effects of selection for elective surgery. However, an effect of the procedure itself cannot be ruled out. Additional studies should attempt to identify the mechanisms that account for the low mortality rate in total hip replacement recipients.
Tables presenting details of comorbidities and the proportional hazards models are available with the electronic versions of this article, on our web site at jbjs.org (go to the article citation and click on “Supplementary Material”) and on our quarterly CD-ROM (call our subscription department, at 781-449-9780, to order the CD-ROM). ▪
In support of their research or preparation of this manuscript, one or more of the authors received grants or outside funding from the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) and from the Arthritis Foundation. None of the authors received payments or other benefits or a commitment or agreement to provide such benefits from a commercial entity. No commercial entity paid or directed, or agreed to pay or direct, any benefits to any research fund, foundation, educational institution, or other charitable or nonprofit organization with which the authors are affiliated or associated.
Investigation performed at Dartmouth Medical School, Lebanon, New Hampshire, and Brigham and Women's Hospital, Boston, Massachusetts
1. , Pukkala E, Pulkkinen P, Visuri T. Causes of death after total hip arthroplasty: a nationwide cohort study with 24,638 patients. J Arthroplasty. 2002;17: 274-81.
2. , Engesaeter LB, Havelin LI, Gjessing HK, Vollset SE. Mortality after total hip replacement: 0-10-year follow-up of 39,543 patients in the Norwegian Arthroplasty Register. Acta Orthop Scand. 2000;71: 19-27.
3. , Pulkkinen P, Turula KB, Paavolainen P, Koskenvuo M. Life expectancy after hip arthroplasty. Case-control study of 1018 cases of primary arthrosis. Acta Orthop Scand. 1994;65: 9-11.
4. . Life expectancy after total hip arthroplasty. J Arthroplasty. 1992;7: 183-6.
5. , Malchau H, Herberts P, Hansson E, Axelsson H, Hansson T. Life expectancy and cost utility after total hip replacement. Clin Orthop Relat Res. 1998;346: 141-51.
6. , Kristensen PW, Petersen MB, Nielsen PT. Patient survival after total knee arthroplasty. 5-year data in 926 patients. Acta Orthop Scand. 1998;69: 35-8.
7. , Barrett J, Katz JN, Liang MH. Total hip arthroplasty: use and select complications in the US Medicare population. Am J Public Health. 1996;86: 70-2.
8. , Steinberg EP, Anderson GF, Herbert R, Hochberg MC. Mortality after elective total hip arthroplasty in elderly Americans. Age, gender, and indication for surgery predict survival. Clin Orthop Relat Res. 1993;295: 119-26.
9. , Albohm MJ, Keating EM, Faris PM, Meding JB. Life expectancy after total hip arthroplasty. J Arthroplasty. 1998;13: 874-5.
10. , Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40: 373-83.
11. , Roos LL, Jollis JG. Adapting a clinical comorbidity index for use with ICD-9-CM administrative data: differing perspectives. J Clin Epidemiol. 1993;46: 1075-90.
12. . Statistical methods for survival data analysis. 2nd ed. New York: Wiley; 1992.
13. , Lemeshow S. Applied survival analysis: regression modeling of time to event data. New York: Wiley; 1999.
14. . Survival analysis using the SAS System: a practical guide. Cary, NC: SAS Institute; 1995.
15. . The status syndrome: how social standing affects our health and longevity. New York: Times Books/Henry Holt; 2004.
16. , Warren JL, Legler JM. Assessing comorbidity using claims data: an overview. Med Care. 2002;40(8 Suppl): IV-26-35.
17. , Bush TL, Penberthy LT. Comorbidity measurement in elderly female breast cancer patients with administrative and medical records data. J Clin Epidemiol. 1997;50: 725-33.
18. , Iwashyna TJ, Christakis NA. The performance of different lookback periods and sources of information for Charlson comorbidity adjustment in Medicare claims. Med Care. 1999;37: 1128-39.
19. , Barrett JA, Katz JN, Phillips CB, Losina E, Lew RA, Guadagnoli E, Harris WH, Poss R, Baron JA. Rates and outcomes of primary and revision total hip replacement in the United States medicare population. J Bone Joint Surg Am. 2003;85: 27-32.