Background/Purpose: Physical activity monitoring studies in rheumatology initially used uniaxial accelerometers.  But advanced triaxial accelerometry technology replaced uniaxial accelerometers. Uniaxial devices measure accelerations in one dimension while triaxial devices measure body accelerations in three planes of movement planes.  The aim of this study is to create a crosswalk table between uniaxial and triaxial physical activity measures to facilitate comparing work from previous studies based on uniaxial devices with new studies using triaxial devices.

Methods: An Osteoarthritis Initiative subset of 185 community dwelling adults aged 45 years or older having knee osteoarthritis or knee osteoarthritis risk factors simultaneously monitored physical activity using both a uniaxial accelerometer (ActiGraph GT1M) and a triaxial accelerometer (ActiGraph GT3X) worn on the waist for 7 days.  Relationship of minute-by-minute output data from the two accelerometers (uniaxial activity counts; triaxial vector magnitude counts) was evaluated using classification tree analysis (Salford Systems CART® v8.0).  Data were split into learning (835,221 matched minutes/130 persons) and test (361,941 matched minutes/55 persons) sets.  The learning set was used to get uniaxial cutpoints and the test set was used to test the cutpoints.  Optimal classification trees (i.e., minimizing misclassification error) identified best-performing cutpoints for uniaxial activity counts to predict triaxial vector magnitude counts using least absolute deviation (LAD) splitting methods.  Medians of triaxial vector magnitude counts in each terminal node were used as predicted values for that node.  Mean absolute deviation (MAD), a measure of model error, was independently estimated in the test set.  Data included only waking hours and excluded sleep periods at night.

Results: 185 participants included 55% with radiographic knee OA (Kellgren-Lawrence grade score ≥ 2), 10% with high pain (SF-12 bodily pain ≥ 3), and 43% with obese weight (body mass index ≥ 30 kg/m²).  Classification tree analyses identified 26 best-performing (i.e., minimum misclassification error) cutpoints for uniaxial activity counts to predict triaxial vector magnitude counts in the learning set (Table 1).  MAD estimate was 302.33 in the test set using 26 optimal cutpoints from the learning set.

Conclusion: This crosswalk table between uniaxial and triaxial physical activity measures will facilitate comparisons of previous uniaxial data with future studies using newer triaxial technology in knee osteoarthritis populations.