Table 3. Within group topic changes over time.

This table presents case (top) and control’s (bottom) derived Dynamic Topic Modeling (Blei et al., 2003) topic correlations, comparing the beginning and end of the treatment interval. We used Rank Biased Overlap (RBO; Amigó et al., 2018) to evaluate differences over time. RBO is a score between 0 (no agreement) to 1 (perfect agreement), representing the overlap between two sets of words and weighting words that are ranked more highly in each list (Bonett and Wright, 2000). We calculated 95 % confidence intervals based on a parametric assumption that the RBO would be distributed similarly to the Kendall Tau, a related measure of agreement (Lapata, 2006). The case and control topics that have asterisks had non-overlapping confidence intervals and are thus significantly different from each other.

Case Rank Biased Overlap
Medication	Intervention	Treatment Goals	Suicide	Treatment Focus
RBO: 0.80 95 % CI [0.75–0.84]	RBO: 0.83* 95 % CI [0.78–0.86]	RBO: 0.73* 95 % CI [0.67–0.78]	RBO:0.73 95 % CI [0.66–0.78]	RBO: 0.83 95 % CI [0.79–0.87]
Control Rank Biased Overlap
Medication	Intervention	Treatment Goals	Suicide	Treatment Focus
RBO: 0.81 95 % CI [0.76–0.85]	RBO: 0.93* 95 % CI [0.91–0.94]	RBO: 0.89* 95 % CI [0.86–0.91]	RBO: 0.79 95 % CI [0.74–0.82]	RBO: 0.88 95 % CI [0.84–0.90]