Recent US government initiatives have led to wide adoption of Electronic Health Records (EHRs). More and more health care institutions are storing patients' data in an electronic format. These EHRs contain valuable information which can be used in important applications like Clinical Decision Support (CDS). So, Information Extraction (IE) from EHRs is a very promising research area. This paper presents a robust method for end-to-end coreference resolution for clinical narratives. For our experiments, we used the datasets provided by i2b2/VA team as part of i2b2/VA 2011 shared task on coreference resolution. One part of this data was annotated according to ODIE guidelines and another part was annotated according to i2b2 guidelines. We designed a global inference strategy for end-to-end coreference resolution which jointly determines the mention types and coreference relations between them. This technique avoids the problem of error-propagation which is common in pipeline systems. For pronominal resolution, we developed different strategies for resolving different pronouns. We report the best results to date on both ODIE and i2b2 data. We got the best results for both types of cases: (1) where gold mentions are already given and (2) for end-to-end coreference resolution. ODIE and i2b2 data are annotated quite differently. Best results on both types of data proves the robustness of our algorithm.