JMIR Publications

ABSTRACT

Background:

Secure instant messaging is increasingly used for inpatient coordination, but evidence from authentic interprofessional message content remains limited, particularly in German-speaking clinical settings.

Objective:

This study aimed to characterize the structural, temporal, lexical, and clinical content patterns of routine messaging and to compare the communicative profiles of direct and group communication.

Methods:

We conducted an observational mixed-methods corpus study of anonymized routine messages cumulatively exported on January 11, 2023, from an on-premises Mattermost platform in clinical production since August 2021. The corpus comprised 7,462 messages from 293 direct-message conversations and 86 group channels. Analyses covered corpus structure, temporal distributions, lexical frequencies and n-grams, direct-versus-group keyness, targeted collocations, communication-unit dispersion, medical terminology, emoji shortcodes, and selected workflow markers. All 4,961 retained anchor-collocate rows were manually assigned to contextual domains; 282 prioritized rows underwent complete, temporally separated intra-rater review. Quantitative results and manual keyword-in-context review were integrated to support the interpretation of recurring communicative profiles.

Results:

The corpus comprised 4,915 direct messages and 2,547 group messages. Group messages contained more cleaned tokens per message than direct messages (35.45 vs. 17.95). Direct messages showed more dialogic and addressive language: du occurred 10.55 versus 1.43 times per 1,000 tokens, and ich 20.04 versus 7.92. Group messages showed more structured patient and workflow markers: hashtagged patient references occurred 36.21 versus 7.90 times per 1,000 tokens, and kein_Todo 7.21 versus 0.12. Friday group messages frequently contained handover (105/631, 16.6%), weekend-preparation (73/631, 11.6%), and no-active-to-do markers (225/631, 35.7%). Unit-level analysis showed that dialogic tokens occurred across most direct conversations, whereas several group workflow markers were concentrated in a small number of functionally specialized channels. Manual validation retained 1,203 clinical entity mentions in direct messages and 2,076 in group messages (13.64 and 22.99 per 1,000 cleaned tokens). Integrated review supported distinct but complementary profiles of bilateral clarification and coordination in direct messages and team-visible handover, task-status, and therapy-process documentation in selected group channels.

Conclusions:

Authentic clinical messaging showed a distinction between broadly distributed bilateral micro-coordination and workflow documentation concentrated in functionally specialized group channels. Secure messaging systems should support modality-specific channel conventions, structured patient referencing, explicit task status communication, and handover-oriented workflows. The transferability of our findings beyond psychosomatic inpatient care requires replication in other coordination-intensive settings.