Predicting long-term trends in inflammatory neuropathy outcome measures using latent class modelling.

Immunoglobulin (Ig) is used to treat chronic inflammatory demyelinating polyradiculoneuropathy (CIDP) and multifocal motor neuropathy with conduction block (MMNCB). Regular infusions may be used for symptom control. Disease activity is monitored with clinical outcome measurements. We examined outcome measure variation during clinically stable periods in Ig-treated CIDP and MMNCB patients. We explored utility of serial outcome measurement in long-term outcome prediction. Retrospective longitudinal analysis of a single neuroscience centre's Ig-treated CIDP and MMNCB patients, 2009-2020, was performed. Mean and percentage change for grip strength, Rasch-built overall disability scales (RODS) and MRC sum scores (MRC-SS) during periods of clinical stability were compared to score-specific minimal clinically important differences (MCID). Latent class mixed modelling (LCMM) was used to identify longitudinal trends and factors influencing long-term outcome. We identified 85 CIDP and 23 MMNCB patients (1423 datapoints; 5635 treatment-months). Group-averaged outcome measures varied little over time. Intra-individual variation exceeded MCID for RODS in 44.2% CIDP and 16.7% MMNCB datapoints, grip strength in 10.6% (CIDP) and 8.8%/27.2% (MMNCB right/left hand) and MRC-SS in 43.5% (CIDP) and 20% (MMNCB). Multivariate LCMM identified subclinical trends towards improvement (32 patients) and deterioration (73 patients) in both cohorts. At baseline, CIDP 'deteriorators' were older than 'improvers' (66.2 vs 57 years, P = .025). No other individual factors predicted categorisation. The best model for 'deteriorator' identification was contiguous sub-MCID decline in more than one outcome measure (CIDP: sensitivity 74%, specificity 59%; MMNCB: sensitivity 73%, specificity 88%). Outcome measure interpretation determines therapeutic decision-making in Ig-dependent neuropathy patients, but intra-individual variation is common, often exceeding MCID. Here we show sub-MCID contiguous changes in more than one outcome measurement are a better predictor of long-term outcome.