Skip to main content

Discussion and conclusions

Your discussion and conclusions sections should answer the question:

What do your results mean?

In other words, the majority of the Discussion and Conclusions sections should be an interpretation of your results. You should:

Discuss your conclusions in order of most to least important.
Compare your results with those from other studies: Are they consistent? If not, discuss possible reasons for the difference.
Mention any inconclusive results and explain them as best you can. You may suggest additional experiments needed to clarify your results.
Briefly describe the limitations of your study to show reviewers and readers that you have considered your study's weaknesses.
Discuss what your results may mean for researchers in the same field as you, researchers in other fields, and the general public. How could your findings be applied?
State how your results extend the findings of previous studies.
If your findings are preliminary, suggest future studies that need to be carried out.
At the end of your discussion and conclusions, state your main conclusions once again.

Example
Source

Feilong Wang, Wenzhi Pan, Shuming Pan, Shuyun Wang, Qinmin Ge and Junbo Ge Usefulness of N-terminal pro-brain natriuretic peptide and C-reactive protein to predict ICU mortality in unselected medical ICU patients: a prospective, observational study. Critical Care 2011;15(1):R42. 

Discussion
In this large scale study of 576 unselected medical ICU patients, we found that NT-proBNP and CRP independently predicted ICU mortality even after adjustment for the APACHE II score and multiple potential confounders including eGFR, age...

...In the present study, we also used a more sensitive test of improvement in model discrimination [27]. We found that the addition of NT-proBNP to the APACHE II score significantly increased the ability to predict ICU mortality as demonstrated by the IDI (6.6%, P = 0.003) and NRI (16.6%, P = 0.007) indices. NT-proBNP was not an independent predictor of ICU mortality in the non-cardiac subgroup after adjustment for APACHE II score and CRP. Kotanidou et al. [13] found that NT-proBNP predicted mortality independently after the adjusted APACHE II score and some inflammatory cytokines levels in non-cardiac ICU patients. But they used TNF-a, IL-6, and IL-10 rather than CRP and enrolled many surgical and multiple trauma cases. In the cardiac subgroup, NT-proBNP independently predicted ICU mortality while the AUC of the APACHE II score was not different from that of NT-proBNP (0.81 ± 0.03 vs 0.77 ± 0.04; P > 0.05). The addition of NT-proBNP to the APACHE-II score can obviously increase predictive ability (IDI = 10.2%, P = 0.018; NRI = 18.5%, P = 0.028). Therefore, although NT-proBNP could predict ICU mortality in unselected medical patents, it appeared to be more useful in cardiac patients than in non-cardiac patients.

...One previous study showed no predictive value of CRP for in-hospital mortality, even in univariate analysis [21]. The scope of the study was rather small (N = 103) and, thus, the statistical power was less than that of our study. Moreover, the endpoint of the previous study was in-hospital mortality but not ICU mortality. The present study revealed that CRP was also an independent predictor of ICU mortality in unselected patients or non-cardiac patients...

Several limitations of our study should be mentioned. First, neither echocardiography was performed nor cardiac function assessed in the present study. The division of subgroups was according to primary admission cause. Thus patients in the non-cardiac group may also have cardiac disease and cardiac dysfunction. However, patients with cardiac diseases as the primary principal diagnosis leading to ICU admission must have cardiac diseases. The statistical conclusion drawn from the cardiac group was appropriate. Second, this was a single-center study, and participants did not include surgery and trauma patients. The value for NT-proBNP in prediction of adverse outcome would be a bit different if the population was different. At last, a limitation of the net reclassification improvement and other reclassification measures is that they depend on the particular categories used [26]. We had used < 10%, 10% to 30%, and 30% to 50%, and > 50% for the risk of ICU death as risk categories. But there are still no well-recognized risk categories now. If the risk categories used had been different, the NRI would be a bit different.

Conclusions
In this large-scale study of unselected ICU patients, we confirmed that NT-proBNP and CRP can serve as moderate independent predictors of ICU mortality. Although the predictive ability was lower compared with the APACHE II score, but the addition of CRP or NT-proBNP or both to the APACHE II score could significantly improve the ability to predict ICU mortality, as demonstrated by IDI and NRI indices. NT-proBNP appeared to be more useful for predicting ICU outcomes in cardiac patients.