Table 2.

Summary of selection process for latent class analysis models

Classes in model	AIC	BIC	SS-BIC	Entropy	VLMR	BLRT	Percentage in each class (in order of size)
							Class 1	Class 2	Class 3	Class 4	Class 5	Class 6
ERAN cohort
1	−5458	−5392	−5738	Not available			100%
2	−5858	−5753	−5823	0.78	2387*	2387*	56%	44%
3	−6229	−6087	−6182	0.81	2951*	2951*	52%	33%	15%
4	−6395	−6214	−6335	0.80	3145**	3145*	48%	27%	13%	12%
5	−6552	−6333	−6480	0.80	3235 ***	3235 *	40%	26%	13%	11%	10%
6	−6645	−6389	−6560	0.82	3311	3311*	40%	26%	11%	11%	10%	2%
British Society for Rheumatology Biologics Registers TNF inhibitor cohort
1	−69,795	−69,695	−69,739	Not available			100%
2	−77,781	−77,623	−77,693	0.65	36,175*	36,175*	62%	38%
3	−79,445	−79,229	−79,325	0.71	38,913**	38,913*	53%	41%	6%
4	−80,851	−80,578	−80,698	0.69	39,753 *	39,753 *	40%	23%	21%	16%
5	−81,896	−81,565	−81,711	0.70	40,464*	40,464*	32%	24%	22%	19%	2%
British Society for Rheumatology Biologics Registers non-biologic cohort
1	−19,209	−19,127	−19,172	Not available			100%
2	−21,497	−21,368	−21,438	0.70	9618*	9618*	56%	44%
3	−22,638	−22,462	−22,557	0.76	10,770*	10,770*	51%	34%	15%
4	−23,170	−22,947	−23,068	0.76	11,349 *	11,349 *	51%	21%	18%	10%
5	−23,457	−23,188	−23,334	0.76	11,623	11,623*	45%	20%	19%	11%	5%

Selection diagnostics for latent class analysis are summarised for models with different numbers of classes. Each model was compared with the previous model with one fewer latent classes. The percentage assigned to each latent class is shown. The model that was selected as the optimal fit is highlighted in bold

Abbreviations: AIC Akaike information criterion, BIC Bayesian information criterion, BLRT Bootstrap likelihood ratio test, ERAN Early Rheumatoid Arthritis Network, TNF Tumour necrosis factor, VLMR Vuong-Lo-Mendell-Rubin likelihood ratio test, SS-BIC Sample size-adjusted Bayesian information criterion

^* p < 0.001

^** p < 0.01

^*** p < 0.05 (significance tests are applied for VLMR and BLRT only)