An Automated, Home-Cage, Video Monitoring-based Mouse Frailty Index Detects Age-associated Morbidity in C57BL/6 and Diversity Outbred Mice

Abstract

Frailty indexes (FIs) provide quantitative measurements of nonspecific health decline and are particularly useful as longitudinal monitors of morbidity in aging studies. For mouse studies, frailty assessments can be taken noninvasively, but they require handling and direct observation that is labor-intensive to the scientist and stress inducing to the animal. Here, we implement, evaluate, and provide a refined digital FI composed entirely of computational analyses of home-cage video and compare it to manually obtained frailty scores in both C57BL/6 and genetically heterogeneous Diversity Outbred mice. We show that the frailty scores assigned by our digital index correlate with both manually obtained frailty scores and chronological age. Thus, we provide an automated tool for frailty assessment that can be collected reproducibly, at scale, without substantial labor cost.

Keywords: 3Rs (reduce replace refine); Age-related pathology; Bioinformatics; Digital biomarkers; Frailty; Home cage.

Figure 1.

Example video stills and methodological details. (A) An example frame of video footage, typical of nighttime video collected in this study. Nighttime illumination was near-IR, resulting in grayscale images. Daytime images were similar, but in RGB color. (B) An example of a modified input image for body-weight prediction error-detection models, with the blue channel modified to indicate masked (blue) versus unmasked (yellow) output from the mouse semantic segmentation model. See Method for details.

Figure 2.

DFI positively correlated with MFI and chronological age in male C57BL/6J mice. (A) Chronological age (x-axis) versus MFI (y-axis). One measurement was taken per mouse in this experiment (see Method), and each dot represents one mouse/measurement. Redundant points are increased in size to indicate the number of data they represent, as indicated in the key. Regression line is in red (R = 0.76; p value = 8.8 × 10⁻⁷). (B) Chronological age (x-axis) versus DFI (y-axis). Up to 3 measurements were taken per mouse in this experiment (see Method), and each dot represents one measurement. Regression line is in red (R = 0.57; p value = 2.8 × 10⁻⁹). (C) MFI (x-axis) versus DFI (y-axis). In this experiment, one MFI measurement was taken per mouse, immediately followed by up to 3 consecutive DFI measurements (see Method). Each dot in this plot represents one DFI measurement, paired with the single MFI measurement from that mouse. Cyan dots are from the first DFI measurement, green dots are from the second, and blue dots are from the third. Regression line is in red (R = 0.48; p value = 1.3 × 10⁻⁶). (D) Correlation between consecutive DFI measurements. From an experiment in which 3 DFI measurements were taken per mouse from consecutive 6-day observation periods (see Method): points compare either the first and second measurements (cyan) or the second and third measurements (green). In both cases, the earlier measurement’s value is plotted on the x-axis and the later’s value on the y-axis. Regression line is in red (R = 0.77; p value = 8.2 × 10⁻¹³). DFI = digital frailty index; MFI = manual frailty index.

Figure 3.

DFI positively correlated with MFI and chronological age in J:DO mice. (A) Chronological age (x-axis) versus MFI (y-axis). Mean (horizontal bars), standard errors (solid vertical bars), and standard deviations (dotted vertical bars) are shown for each of 3 measurement instances taken for each of 8 birth cohorts (with cohort 8 split into 2 subgroups: see Key). The linear regression of age versus MFI for all individual values is shown in green. (B) Chronological age (x-axis) versus DFI (y-axis). Plotted as in (A). (C) MFI (x-axis) versus DFI (y-axis). For each of 3 MFI/DFI measurements taken for each cohort, a cross is drawn whose center is the mean value on each axis for one cohort/measurement, and whose bars indicate the standard errors of those means along each axis. Colored according to the Key. The linear regression of MFI versus DFI for all individual values is shown in green. (D) A path diagram depicting a model for consideration of the possible relationships between MFI, DFI, and chronological age. See Results and Discussion. (E)Age-normalized MFI (x-axis) versus age-normalized DFI (y-axis), plotted as in (D). (F) Chronological age (x-axis) versus combined frailty index (CFI: the average of MFI and DFI; y-axis). Plotted as in (A). DFI = digital frailty index; MFI = manual frailty index.

Figure 4.

Most DFI components positively correlated with MFI and chronological age. (A) A heat map indicating the correlation coefficients between individual components of the DFI (and, at bottom, the full DFI) versus MFI and chronological age, among C57BL/6J male mice. (B) A heat map indicating the correlation coefficients between individual components of the DFI (and, at bottom, the full DFI) versus MFI and chronological age, among J:DO mice of both sexes. (C) Hierarchical clustering (left) of individual MFI and DFI components, and chronological age, based on the correlation coefficients depicted in the heatmap (right). MFI components are labeled in orange; DFI components are labeled in blue; chronological age is labeled in green. Only MFI components observed with non-zero values >5 times across the study are included. (D) Heat maps indicating the correlation coefficients between individual components of the DFI versus MFI and chronological age, among J:DO mice, independently analyzed for each sex. (E) Heat maps indicating the correlation coefficients between individual components of the MFI (those included in panel C) versus DFI and chronological age, among J:DO mice, independently analyzed for each sex. DFI = digital frailty index; MFI = manual frailty index.