In this study, we found that MHD patients performed worse in the St1, St3 and St5 subtasks of the TUG test compared with healthy controls, while the differences in St2 and St4 were not significant. The St3 subtask shown largest difference, St1 and St5 followed, with Cohen's d effect sizes of 1.378, 1.112, and 0.801, respectively. Namely, MHD patients had a worse performance in turning around task followed by standing up and sitting down task. Turn durations in the TUG test were significantly associated with balance function[15], and time taken to accomplish the individual 180-turn test was highly correlated with the Berg balance scale[16]. Sit-to-stand is a common way to assess lower extremity strength, for example, the time taken to accomplish 5STS was significantly associated with the muscle strength of knee flexors reported by previous studies[17, 18]. Hence, from a basic motor function perspective, MHD patients may have a greater decline in balance function followed by lower extremity strength. Targeted exercise training can improve motor performance, reduce symptoms such as osteoporosis and muscle atrophy, and enhance quality of life for MHD patients [19]. However, impairments in different dimensions of motor function were varying. Pajek et al. quantified deficits in multiple motor abilities using several motor tests simultaneously, and found that MHD patients had largest functional deficits in balance, flexibility, and lower extremity strength [20], which was consistent with our findings. Determining appropriate rehabilitation goals and developing optimal training programs are particularly important for exercise interventions, but it is impractical to use multiple instruments simultaneously to assess motor function in a clinical setting. The TUG subtask durations could be used to quantify the dysfunctions from multiple motor dimensions, which may help to determine the motor rehabilitation goals via a single test.
In the MHD group, we performed an exploratory regression analysis of features that may be associated with St1-St5 durations and the TUG total time, we found that the coexistence of DM may associated with the prolonged St3 duration. This suggested that MHD patients with DM may have higher risk of impaired balance function and deserve further attention. The individual effects of renal disease, dialysis, and diabetes on motor function have been reported in numerous studies, while fewer studies have been done on MHD patients with diabetes. The first research on gait and balance in MHD patients with diabetes was reported by Zhou et al [21]. They collected kinematic parameters of patients walking 15m by IMUs on bilateral ankles and a significant decrease was found on gait and balance in DM patients undergoing dialysis, compared to patients with DM alone. Our results could be complementary to the above study, in that we found a potential effect of DM on balance function among MHD population. Shirai et al. compared balance function in dialysis and non-dialysis chronic kidney disease patients using a force platform and found that DM factor was associated with longer length of the center of pressure during static standing, but not associated with the TUG total time. Our finding was consistent with the above-mentioned study, suggesting that balance function was worse in MHD patients with DM, which may be due to chronic neuropathy in diabetes and the associated cognitive, motor, and sensory dysfunction. Compared to the TUG total time, St3 duration may have higher sensitivity in balance evaluation, and St3 duration was expected to be a predictor of falls for MHD patients with DM. Since this analysis is a post hoc exploratory analysis, the evidence of regression results still needs to be improved, and further separate studies on TUG subtask performance in MHD patients with DM are still needed.
The CV based TUG time segmentation established in this study has good accuracy, could faithfully reflect the time taken to complete each subtask, and the result was more objective as algorithm would always adhere a same rule to locate transition points. ICC and Bland-Altman agreements analysis showed good agreements between the durations extracted by CV and the ground truths. The ICC values were higher than 0.75 for all subtasks that means the subtask durations based on the CV can replace the repeated manual observations. The 95% limits of agreement were within 0.6 seconds, which was much lower than the clinically significant difference of TUG total time defined in previous research [22], indicating that measurement errors of this method would not affect interpretation of the TUG total time changes. The accurate skeleton key points were the prerequisite for our study, we chose Mediapipe as the key points marking solution because it spends fewer computing resources for single person detection, can run on mobile processors, and can be deployed on mobile phones, laptops, and web[12]. These advantages could facilitate the reproduction of our method, this test would be conducted via internet on computer or mobile devices from anywhere in the world, after we deploy the algorithm to the cloud platform. CV based marker-less human motion analysis has a broad application prospect in the field of telemedicine especially during the COVID-19 epidemic [23], when we are unable to perform a face-to-face motor assessment, it can give us a relatively reliable result, and the implementation of automatic TUG test is only a small aspect of the CV application.
The TUG subtask durations had good correlations with the kinematic parameters of individual motor tests. The subtask durations of TUG test could replace individual motor tests to some extent and show promise to simplify clinical functional assessment protocols. St1 and St5 durations could be used to evaluate sitting and standing performance, respectively. Unlike the 6-stage segmentation [24], we divided both the final turning and sitting down motions into St5 phase. Weiss et al. found that there were two transition strategies at St5 phase: the distinct transition strategy (DTS), in which turning and sitting down motion were performed in sequence; overlapping transition strategy (OTS), in which, turning motion have not finished before the sitting motion started [25]. Hence, the 6-stage method may be inappropriate for patients employing OTS strategy, let alone to compare with healthy controls. Although St5 contains 2 motions, it mainly reflects the sitting down function as St5 duration showing the largest correlation with the sitting time of 5STS (r = 0.704). Both St2 and St4 were walking phases, but St2 had a higher correlation with the gait speed of 2MWT, St2 could better represents the patient's walking ability consequently. A shorter walking distance in St4 may explain this difference, as there were two turning tasks on both end of St4 and leading to a shorter period (2.54 vs. 2.94s). In summary, St1, St2, St3, and St5 were good indicators of standing up, walking, balancing, and sitting down functions, respectively, and these functions can be assessed in a single test.
Our study has several limitations. First, to ensure the safety and compliance of the test, we excluded the participants unable to complete tests independently. So, there was a possibility of selection bias toward patients of less motor deficit in our sample, and further studies were needed to quantify performance of patients completing the test with help of others or assistive devices. Secondly, though this study chose a 3D key point marking model, only temporal parameters of the TUG test were extracted, some previous studies have extracted certain spatial gait parameter successfully from walking videos[26], accurate temporal segmentation of the TUG test is a prerequisite for spatial parameter extraction, works on the kinematic characteristics of MHD patients in each subtask will be done in the further.