Title: Evaluation of patient related factors influencing outcomes after total hip replacement

On May 29:th I successfully defended my thesis at the Karolinska Institute and I’m now a “Doctor of Philosophy“, i.e. PhD. It has been a fun and rewarding project that spurred me into starting this blog and diving into R. Below you can find the thesis abstract and my reflections on the subject.

Background

Basic patient factors such as age, sex, and co-morbidities are poorly understood in the context of re-operation rates and patient-reported outcome measures after total hip replacement (THR) surgery.

Aims

The aims of this thesis were to investigate:

1. If co-morbidity measures developed for mortality outcomes also are applicable for re-operations.
2. How age influences health-related quality of life (HRQoL).
3. If sex/age interacts with self-administered Charnley classification in regards to HRQoL.
4. The generalizability of HRQOL estimates.
5. If antidepressant use impacts patient-reported outcomes.

Methods

All studies were based upon the Swedish Hip Arthroplasty Register (SHAR) database. For study I, data from SHAR was cross-matched with the National Patient Register through which 3 different co-morbidity measures were calculated: the Charlson score, the Elixhauser score, and the Royal College of Surgeons (RCS) Charlson’s score. The three scores were then compared using survival analysis with implant re-operation performed between 0 to 2 years and 2 to 12 years.

In study II and III we used the SHAR’s PROM database with HRQoL outcomes as measured by EQ-5D and EQ-VAS. In study II we modeled age using linear regression in combination with restricted cubic splines in order to study the relationship between age and HRQoL. In study III we used linear regression with interaction terms evaluated by ANOVA-tests, subset, and EQ-5D dimension specific analyses.

In study IV we linked the SHAR’s PROM database to the National Patient Register and a Danish cross-sectional sample. The Charlson co-morbidity measure was calculated as in study I, and effect modification by country was investigated through terms of interaction, evaluated as in study III.

In study V, we cross-matched the SHAR’s PROM database with the Prescribed Drug Register. We calculated the usage of antidepressants using regular expressions. Measures for compliance, treatment change, and indication were retrieved from the prescription text.

Results

Study I

0-2 years, only the Elixhauser score showed significant risk increase with increased score for both 1-2 and ≥ 3 co-morbidities. The predictive C-statistic in this period for the Elixhauser score was poor, 0.52. None of the measures proved to be of any value between 2-12 years.

Study II

Both the EQ-5D index and EQ VAS exhibited a non-linear relationship with age, they were fairly unaffected by age until the patient’s late sixties, after which it had a negative impact.

Study III

We found that women in category C had a poorer EQ-5D outcome compared to men. This effect was mostly due to the fact that women failed to improve in the mobility dimension, only 40% improved, while 50% of men improved. Age did not interact with Charnley class. We also found that the classification performed best without splitting or aggregating classes.

Study IV

Danish patients had an overall higher EQ-5D index and EQ VAS than Swedish patients (p-value < 0.001). After regression analysis, the estimated coefficients for sex, age, or the Charlson score did not differ between countries for either the EQ-5D index (p-value = 0.83) or EQ VAS (p-value = 0.41) one year after THR.

Study V

Antidepressants were used by 9% of the cases (n = 954). Patients using antidepressants had poorer HRQoL, more pain, and experienced less satisfaction. Preoperative antidepressant use was independent of patient-reported anxiety/depression in predicting PROs one year after THR. Discontinuation of treatment was negatively associated with pain and satisfaction at one year.

Conclusions

Study I

We failed to validate any of the scores for re-operations after total hip arthroplasties, although the Elixhauser score may be useful for estimating the co-morbidities relevant to the risk of re-operation within 2 years. The co-morbidity associated risk increase was small, and is undoubtedly best suited to the study of large samples and not individual patients.

Study II

There is a non-linear relationship for age and HRQoL in patients receiving THR; resulting in residual confounding if treated as a simple linear term or categorically in the regression. The implication of this is important, as age is a common confounder. The same applies to the preoperative EQ-5D index and EQ VAS.

Study III

The self-administered Charnley classification is a reliable instrument with interesting properties easy to utilize in everyday clinical practice. There is also strong evidence that women in Charnley class C fail to improve their mobility as much as men.

Study IV

There are clear similarities in how basic predictors influence patient-reported outcomes in patients with THR in Sweden and Denmark. Apparent cultural, social, and other such differences among these countries are not reflected in these predictors.

Study V

Antidepressants have a negative influence on patient-reported outcomes 1 year after THR, independent of the pre-operative EQ-5D anxiety/depression dimension. We also found that discontinuation of treatment prior to surgery is associated with poorer outcomes in the dimensions of pain and satisfaction.

My reflections

I clearly underestimated at start the difficulty in finding a predictive score for re-operations. The required data for adjusting patient case-mix is most likely not located in disease codes and with hindsight – not that surprising. Despite this, the negative finding is still interesting as many believe that adjusting the results using a co-morbidity score takes care of confounding.

I also had a lot of fun with the age study. I believe that nature is not nice and simple and that splines are an amazing tool for dealing with non-linearity. I’ve previously encountered arguments such as this being too complicated for regular doctors, to this my response is:

• I didn’t encounter any resistance from either reviewers or editors.
• My colleagues not trained in statistics found this easier to understand than the traditional linear or categorical relations.

I also liked the Charnley class article; the results are plain, simple, and relevant. Combined with the transitionplot (in my Gmisc-package) this is my favorite article.