One of the independent predictor identified for inhospital complication is platelet count. It is not categorized, inserted in regression as continuous variable. Can any one tell me how to interprete the hazard ratio?
HR=1.0002 p value=0.039
Or, to make that a bit easier to digest:
under the applied model, the data suggests that for every 100 platelets more, the hazard increases by 2% (->HR). This is an estimate, an it is uncertain. There is no indication of the uncertainty given, but the p-value indicates that the data is sufficient to at least trust the sign of the change in the hazard (more platelets -> hiher hazard).
If this result has some practical meaning depends on the experimental design (how were patients selected?, in what hospital?, was the treatment here "typical"?...) and if the model is appropriate (are the assumptinos of the model resonable?, are there relevant covariates, and if so, are they considered?, is the functional part of the model resonable [e.g., if the model describes a linear change: can the relationship be assumed linear at least over the typical range of platelet counts?]).
For an interpretation it would also be nice to have the confidence interval of HR. The p-value only tells us that the lower limit of this interval ist a tiny bit larger than 1 (the harard are almost equal, just a tiny bit larger when there are more platelets), but the upepr limit is missing. This would possible be required to judge if the data is or is not compatible with a clinically relevant change in the HR. For instance, if an increase by 100 platelets would lead to less than a 10% increase in the hazard, but the upper limit of the confidence interval is 1.004 (=4% increase per 100 more platelets), the data would suggest that there is a tiny positive effect (HR>1), but this is so small that we should not think that it can be clinically relevant.
Increase of one additional platelet increases the event risk (inhospital complication?) by a factor of 1.0002.
Or, to make that a bit easier to digest:
under the applied model, the data suggests that for every 100 platelets more, the hazard increases by 2% (->HR). This is an estimate, an it is uncertain. There is no indication of the uncertainty given, but the p-value indicates that the data is sufficient to at least trust the sign of the change in the hazard (more platelets -> hiher hazard).
If this result has some practical meaning depends on the experimental design (how were patients selected?, in what hospital?, was the treatment here "typical"?...) and if the model is appropriate (are the assumptinos of the model resonable?, are there relevant covariates, and if so, are they considered?, is the functional part of the model resonable [e.g., if the model describes a linear change: can the relationship be assumed linear at least over the typical range of platelet counts?]).
For an interpretation it would also be nice to have the confidence interval of HR. The p-value only tells us that the lower limit of this interval ist a tiny bit larger than 1 (the harard are almost equal, just a tiny bit larger when there are more platelets), but the upepr limit is missing. This would possible be required to judge if the data is or is not compatible with a clinically relevant change in the HR. For instance, if an increase by 100 platelets would lead to less than a 10% increase in the hazard, but the upper limit of the confidence interval is 1.004 (=4% increase per 100 more platelets), the data would suggest that there is a tiny positive effect (HR>1), but this is so small that we should not think that it can be clinically relevant.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression