10.08.2013 Views

Het volume van chirurgische ingrepen en de impact ervan op ... - KCE

Het volume van chirurgische ingrepen en de impact ervan op ... - KCE

Het volume van chirurgische ingrepen en de impact ervan op ... - KCE

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

216 Volume Outcome <strong>KCE</strong> reports 113<br />

• In cardiology, the Belgian Working Group for Interv<strong>en</strong>tional Cardiology<br />

(BWGIC) disposes of data for CABG and PCI, and, although we did not<br />

link MCD data to BWGIC registry due to time constraints, this type of<br />

linkage has already be<strong>en</strong> performed in the past and could be redone.<br />

• For orth<strong>op</strong>aedic procedures, only very rec<strong>en</strong>t the National Institute for<br />

Health and Disability Insurance (NIHDI) started the electronic registry<br />

ORTHOpedic (Prosthesis Id<strong>en</strong>tification Data ORTHOpri<strong>de</strong>) which will<br />

inclu<strong>de</strong> all prostheses of hip and knee. 343 Such initiatives will <strong>de</strong>finitely be<br />

of b<strong>en</strong>efit to the quality of future studies.<br />

We cannot rule out confounding by unmeasured characteristics of pati<strong>en</strong>ts in our study.<br />

Nevertheless, we do not believe that limitations related to risk adjustm<strong>en</strong>t threat<strong>en</strong> our main<br />

conclusions about the association betwe<strong>en</strong> <strong>volume</strong> and outcome.<br />

5. Appr<strong>op</strong>riate statistical methods should be used. Regression mo<strong>de</strong>ls are<br />

available that respect the hierarchical nature of the data (pati<strong>en</strong>ts nested<br />

within surgeons, surgeons nested within c<strong>en</strong>tres), and account for the<br />

correlations within these clusters. The failure to inclu<strong>de</strong> any type of<br />

adjustm<strong>en</strong>t for those correlations would lead to falsely high statistically<br />

significant effects. G<strong>en</strong>eralized Estimating Equations (GEE) and G<strong>en</strong>eralized<br />

Linear Mixed Mo<strong>de</strong>ls (GLMM) and are two examples of methods that can be<br />

used to analyze hierarchical data.<br />

6. The graphical pres<strong>en</strong>tation of results. The funnel plot is a good and “easy to<br />

use” tool. It avoids spurious ranking of institutions, spurious stigmatization of<br />

low <strong>volume</strong> c<strong>en</strong>tres, and allows for an informal assessm<strong>en</strong>t of any <strong>volume</strong><br />

outcome relationship.<br />

7. S<strong>en</strong>sitivity analyses and robustness of the results. Results should be<br />

transpar<strong>en</strong>t. In our study, effects of <strong>volume</strong> were always pres<strong>en</strong>ted with and<br />

without adjustm<strong>en</strong>t for case mix (based on administrative data only or based<br />

on all clinical data available) so that one can assess the influ<strong>en</strong>ce of case mix<br />

on the <strong>volume</strong> effect. S<strong>en</strong>sitivity analyses were also performed wh<strong>en</strong> there<br />

were huge <strong>volume</strong> outliers, where it is difficult to differ<strong>en</strong>tiate the effect of<br />

the <strong>volume</strong> or other characteristics of that c<strong>en</strong>tre.<br />

8. Problem of missing data on cancer stage: on average 30% missing data,<br />

sometimes 40% in small <strong>volume</strong> hospitals. The fact that mortality rate was<br />

not substantially higher in the pati<strong>en</strong>ts whose stage was missing in the BCR<br />

seems to indicate, that, in this study, these pati<strong>en</strong>ts are randomly divi<strong>de</strong>d into<br />

the four disease stages. I<strong>de</strong>ally, though, this assumption should have be<strong>en</strong><br />

checked with the help of s<strong>en</strong>sitivity analyses, but this was not done due to<br />

time constraints.<br />

In addition, we noticed that many hospitals – low-<strong>volume</strong> as well as high<strong>volume</strong><br />

– missed data on stage and that the perc<strong>en</strong>tage of missing data varied<br />

among these hospitals. Despite the failure to retrieve information on disease<br />

stage, this problem did not restrain us from drawing conclusions on the<br />

<strong>volume</strong> outcome association. Nevertheless, this problem of missing data on<br />

disease stage (and other variables useful for risk adjustm<strong>en</strong>t) supports the<br />

need for complete and accurate data collection.<br />

9. Sample size is sometimes not suffici<strong>en</strong>t in one year: analysis of several years<br />

for rare tumours or procedures (pancreas, oes<strong>op</strong>hagus, heart transplant) is<br />

required.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!