Determination and application of cut points for accelerometer-based activity counts of activities with differing intensity in pet dogs

Kathryn E. Michel Department of Clinical Studies-Philadelphia, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA 19104.

Search for other papers by Kathryn E. Michel in
Current site
Google Scholar
PubMed
Close
 DVM, MS
and
Dorothy Cimino Brown Department of Clinical Studies-Philadelphia, School of Veterinary Medicine, University of Pennsylvania, Philadelphia, PA 19104.
Center for Clinical Epidemiology & Biostatistics, School of Medicine, University of Pennsylvania, Philadelphia, PA 19104.

Search for other papers by Dorothy Cimino Brown in
Current site
Google Scholar
PubMed
Close
 DVM, MSCE
Full access

Abstract

Objective—To investigate whether an accelerometer-based activity monitor could be used in pet dogs to differentiate among and delineate the amount of time spent in activities of differing intensity.

Animals—104 dogs.

Procedures—For the first phase of the study, each dog (n = 104) wore an accelerometer-based activity monitor and was led through a series of standard activities (recumbency [sedentary], walking, and trotting). Receiver operating characteristic curves were generated to determine the optimal activity counts for predicting whether a dog was sedentary, walking, or trotting. For the second phase of the study, dogs (n = 99) wore an activity monitor on their collars continuously for 14 days at home; intensity of activity for each dog was classified by use of cut points determined on the basis of results obtained during the first phase of the study.

Results—Analysis of receiver operating characteristic curves indicated that there was 100% specificity and 100% sensitivity in distinguishing sedentary activity from walking activity and 92% specificity and 92% sensitivity in distinguishing trotting activity from walking activity. Analysis of data collected during the 14-day period at home indicated that dogs were sedentary most of the time (median, 87%; range, 65% to 95%).

Conclusions and Clinical Relevance—Counts recorded by an accelerometer-based activity monitor could be used to discriminate effectively among standardized activities in pet dogs. There is potential for use of the method to improve the ability of clinicians and researchers to accurately estimate a pet dog's daily energy requirement.

Abstract

Objective—To investigate whether an accelerometer-based activity monitor could be used in pet dogs to differentiate among and delineate the amount of time spent in activities of differing intensity.

Animals—104 dogs.

Procedures—For the first phase of the study, each dog (n = 104) wore an accelerometer-based activity monitor and was led through a series of standard activities (recumbency [sedentary], walking, and trotting). Receiver operating characteristic curves were generated to determine the optimal activity counts for predicting whether a dog was sedentary, walking, or trotting. For the second phase of the study, dogs (n = 99) wore an activity monitor on their collars continuously for 14 days at home; intensity of activity for each dog was classified by use of cut points determined on the basis of results obtained during the first phase of the study.

Results—Analysis of receiver operating characteristic curves indicated that there was 100% specificity and 100% sensitivity in distinguishing sedentary activity from walking activity and 92% specificity and 92% sensitivity in distinguishing trotting activity from walking activity. Analysis of data collected during the 14-day period at home indicated that dogs were sedentary most of the time (median, 87%; range, 65% to 95%).

Conclusions and Clinical Relevance—Counts recorded by an accelerometer-based activity monitor could be used to discriminate effectively among standardized activities in pet dogs. There is potential for use of the method to improve the ability of clinicians and researchers to accurately estimate a pet dog's daily energy requirement.

The daily maintenance energy requirement of a healthy adult dog includes the metabolic demands of basic life processes and energy expended for thermo-regulation, assimilation of nutrients from the diet, and physical activity. This last component of daily energy expenditure, voluntary physical activity, can vary considerably among individual animals. Generally, feeding directions on pet food labels or recommendations provided by veterinary health professionals are determined on the basis of predictive equations. These equations have been derived directly or indirectly from data collected by use of kenneled dogs in research conditions. Currently, there are only preliminary data quantifying the energy expenditure of free-living pet dogs; therefore, the equations currently in use for predicting energy requirements of dogs have not been validated for the purpose for which they are most commonly used.

Pet dogs generally are housed in environments that differ considerably from those of kenneled dogs. In addition, the environments of pet dogs can vary substantially among households. Furthermore, the daily routine of kenneled dogs typically is uniform regardless of the day of the week, which is in contrast to that of pet dogs, which may vary from day to day and can be influenced by the activities of their owners. Also, the kinds of activities in which pet dogs participate are likely to be much more variable and uncontrolled, compared with those of dogs in a laboratory setting. The ability to quantify activity could lead to a better understanding of the energy expenditure of pet dogs and the validity of the currently used predictive equations.

An accelerometer-based device that can continuously measure the intensity, frequency, and duration of movement for extended periods has been investigated for use in monitoring the activity of pet dogs. In another study1 conducted by our research group, we found that there is no significant impact of signalment or body conformation on the mean activity counts recorded by this monitor when the activity is adequately controlled and all dogs perform the same movements. Furthermore, data collection for a period of 7 days provides relatively stable estimates of activity in dogs and includes the days (ie, weekends) with the highest potential for changes in activity.2

The objectives of the study reported here were to investigate whether an activity monitor could be used to differentiate the intensity of activity of pet dogs through the establishment of cut points for activity counts (eg, optimal activity counts) that could be used to discriminate among sedentary, light, and moderate to vigorous activities and to use these cut points to determine the percentage of time that pet dogs spend in activities of differing intensity.

Materials and Methods

Animals—Pet dogs belonging to staff and students of the School of Veterinary Medicine at the University of Pennsylvania were recruited to participate in the study. On the basis of medical history and physical examination, all dogs enrolled were judged to be free of clinically important orthopedic or neurologic disease to ensure that they would be able to perform the prescribed controlled activities without difficulty or discomfort. To ensure inclusion of a diverse cohort, at least 15 dogs were recruited in each of 5 weight ranges (< 10 kg, 10 to 20 kg, 21 to 30 kg, 31 to 40 kg, and > 40 kg). Signalment data, including age, sex, body weight, and body condition score (scale of 1 to 9), were recorded for each dog. The protocol for the study was approved by an institutional animal care and use committee.

Experimental procedures—Activity in each dog was monitored by use of an omnidirectional accelerometer-based devicea that continuously measured the intensity, frequency, and duration of movement. A detailed description of the monitor and its mechanism of action have been reported elsewhere.3 The activity monitor, which was approximately 29 × 37 mm, was attached to the collar of each dog and positioned ventrally on the neck. The study was conducted in 2 phases.

Phase 1—To establish the cut points for sedentary, light, and moderate to vigorous activity, each dog was guided through a series of controlled activities. Data were collected for 3 minutes for each activity. Prior to beginning the 3-minute data collection period, each dog was acclimated by engaging it in the standard activity until the dog appeared comfortable. On completion of data collection for an activity, the dog was given sufficient time to completely recover before proceeding to the next activity. The heart rate of the dog was measured immediately before each activity commenced to confirm a return to the baseline status.

Each dog first performed a sedentary activity. Each dog remained in sternal or lateral recumbency for the entire data collection period. The second activity was walking. The dogs were walked on a leash on a designated smooth, level path. The third activity was trotting. The dogs were trotted on a leash along a designated smooth, level path. The accelerometer epoch (period during which activity counts were accumulated before the sample was saved) was set at 15 seconds to allow maximum data collection over the 3-minute activity period.

Phase 2—An activity monitor was placed on the collar of each of the pet dogs. Data were continuously collected for 14 days while dogs were in their home environment engaged in their typical activities. The accelerometer epoch was set at 1 minute to permit data collection over an extended period.

Statistical analysis—Descriptive statistics were calculated. Continuous data were expressed as mean and SD for normally distributed data or median and range for nonnormally distributed data. Categorical data were expressed as frequencies. The Wilcoxon signed rank test was used to compare counts for sedentary activity with counts for walking activity and counts for walking activity with counts for trotting activity. Then, ROC curves were generated to determine the optimal activity counts for 15-second epochs used to predict whether a dog was sedentary, walking, or trotting. An ROC curve represents the relationship between the true-positive rate (the probability of classifying a trotting dog as trotting [ie, sensitivity]) and the false-positive rate (the probability of classifying a walking dog as trotting [ie, 1 − specificity]) for each possible cut point. The AUC for an ROC curve is a global summary statistic of the predictive value across all possible cut points. This allows classification of a test as highly accurate (0.9 < AUC < 1), moderately accurate (0.7 < AUC ≤ 0.9), less accurate (0.5 < AUC ≤ 0.7), or noninformative (AUC ≤ 0.5).4 Counts for 15-second epochs then were converted to counts for 1-minute epochs, and all analyses were repeated.

For phase 2, a dog's intensity of activity for each minute of recording was classified by use of the total counts for that minute and the cut points determined on the basis of results obtained during phase 1 of the study. The percentage of time dogs spent in sedentary, light, or moderate to vigorous activity each day was calculated and evaluated for the 14 days of data collection. The percentages of time dogs spent in sedentary, light, or moderate to vigorous activities each day were reported as median and range values. Linear regression analysis was used to determine the association of signalment on the amount of time dogs spent in activities of differing intensity.

All analyses were performed by use of commercially available software.b For all analyses, values of P < 0.05 were considered significant.

Results

Phase 1—Pet dogs (n = 104) were included in the analysis for determination of cut points for activities of differing intensity. Body weight of the dogs was < 10 kg [n = 22], 10 to 20 kg [20], 21 to 30 kg [23], 31 to 40 kg [21], and > 40 kg [18]). Mean ± SD body weight was 25 ± 13 kg (range, 4 to 62 kg), and median body condition score was 5.5 (range, 4 to 8). The mean age was 4.9 ± 3.2 years (range, 1 to 14 years). Fifty-eight (56%) dogs were males, and 46 (44%) dogs were females.

Activity counts for 15-second epochs were evaluated (Figure 1). Counts for sedentary activity (median, 0; range, 0 to 42) differed significantly (P < 0.001) from the counts for walking activity (median, 287; range, 102 to 660), and counts for walking activity differed significantly (P < 0.001) from the counts for trotting activity (median, 776; range, 340 to 1,519). Analysis of the ROC curves indicated that the activity count was highly accurate for use in differentiating sedentary activity from walking activity. The AUC for this ROC curve was 1.00. At an activity count of 42, there was 100% specificity and 100% sensitivity in distinguishing sedentary activity from walking activity. Analysis of the ROC curves also indicated the activity count was highly accurate for use in differentiating between walking and trotting activities (Figure 2). The AUC for this ROC curve was 0.9813 (95% confidence interval, 0.9750 to 0.9965). At an activity count of 429, there was 92% specificity and 92% sensitivity in distinguishing walking activity from trotting activity.

Figure 1—
Figure 1—

Box-and-whiskers plots of activity counts recorded by an accelerometer-based activity monitor that collected data at 15-second epochs for 104 dogs that performed each of 3 controlled activities (recumbency [sedentary], walking, and trotting). Each box represents the lower and upper quartiles of the data, the horizontal line in each box represents the median, and the whiskers indicate the smallest and largest values that are not outliers (outliers are represented as black circles).

Citation: American Journal of Veterinary Research 72, 7; 10.2460/ajvr.72.7.866

Figure 2—
Figure 2—

An ROC curve for describing the sensitivity and specificity for activity counts recorded by an activity monitor during 15-second epochs in 104 dogs to differentiate walking activity from trotting activity. At an activity count of 429, there is 92% specificity and 92% sensitivity in distinguishing walking activity from trotting activity. The AUC for the ROC curve is 0.9813.

Citation: American Journal of Veterinary Research 72, 7; 10.2460/ajvr.72.7.866

Data then were converted to activity counts for 1-minute epochs, which resulted in the following activity counts: median sedentary activity, 20 (range, 0 to 221), median walking activity, 1,196 (range, 414 to 2,475), and median trotting activity, 3,027 (range, 1,340 to 6,067). Counts for the sedentary and walking activities differed significantly (P < 0.001). Similarly, counts for the walking and trotting activities also differed significantly (P < 0.001). Analysis of the ROC curves indicated that the activity count was highly accurate for use in differentiating sedentary activity from walking activity. The AUC for this ROC curve was 1.00. At an activity count of 204, there was 100% specificity and 100% sensitivity in distinguishing sedentary activity from walking activity. Analysis of the ROC curves indicated that the activity count also was highly accurate for use in differentiating walking activity from trotting activity. The AUC for the ROC curve was 0.9813 (95% confidence interval, 0.9750 to 0.9965). At an activity count of 1,751, there was 92% specificity and 92% sensitivity in distinguishing walking activity from trotting activity.

Phase 2—Pet dogs (n = 99) were included in the analysis to determine the amount of time spent in activities of differing intensity over a 14-day interval. Body weight of the dogs was < 10 kg [n = 20], 10 to 20 kg [20], 21 to 30 kg [22], 31 to 40 kg [21], and > 40 kg [16]). Mean ± SD body weight was 24.9 ± 14.1 kg (range, 2.3 to 71.0 kg). Median age was 3.5 years (range, 1 to 12 years). Fifty-five (56%) dogs were males, and 44 (44%) dogs were females.

The percentages of time dogs were engaged in sedentary, light, or moderate to vigorous activities for each day were calculated (Table 1). The median percentage of time that dogs were engaged in sedentary, light, or moderate to vigorous activity for the entire 14 days was 87% (range, 65% to 95%), 11% (range, 4% to 31%), and 2% (range, 0% to 13%), respectively. The distribution of the median percentage of time spent in activities of differing intensity for the 14 days for each dog was determined (Figure 3). On the basis of linear regression analysis, the only variable significantly associated with the amount of time spent in activities of differing intensity was age of the dog. For every 1-year increase in age, there was a 0.9% increase in the amount of time dogs were sedentary and 0.7% and 0.2% decreases in the amount of time dogs spent in light and moderate to vigorous activities, respectively. The age-adjusted median percentage of time spent in activities of differing intensity was calculated for dogs that were 1, 7, and 12 years old (Table 2).

Figure 3—
Figure 3—

Distribution of the amount of time spent by dogs (n = 99) in their home environments performing activities of differing intensity during a 14-day period.

Citation: American Journal of Veterinary Research 72, 7; 10.2460/ajvr.72.7.866

Table 1—

Median (range) percentage of time spent in activities of differing intensity each day during continuous monitoring of activity of 99 healthy pet dogs in their home environments for 14 days.

 Week 1Week 2
Activity intensitySatSunMonTueWedThuFriSatSunMonTueWedThuFri
Sedentary86 (65–95)85 (59–96)87 (64–95)87 (68–98)87 (66–96)87 (62–98)87 (60–97)86 (63–96)86 (59–97)88 (70–96)88 (64–96)88 (65–96)88 (68–97)88 (68–97)
Light12 (5–34)12 (4–38)10 (5–33)11 (3–30)11 (4–29)10 (3–40)10 (3–34)13 (5–34)12 (3–42)10 (4–30)10 (4–35)10 (4–32)11 (3–39)10 (3–30)
Moderate to vigorous2 (0–18)2 (0–33)2 (0–17)2 (0–10)2 (0–8)2 (0–9)2 (0–15)2 (0–10)2 (0–12)2 (0–20)2 (0–16)2 (0–13)2 (0–14)2 (0–12)

Values in a column may not sum to 100% because of rounding.

Table 2—

Age-adjusted median percentage of time spent in activities of differing intensity for dogs of various ages, as calculated on the basis of linear regression analysis.

Activity intensity1 year old7 years old12 years old
Sedentary838892
Light14107
Moderate to vigorous321

Discussion

In the study reported here, we found that the counts recorded by an accelerometer-based activity monitor could be used to discriminate effectively among standardized activities in pet dogs. Furthermore, analysis of the data indicated that the pet dogs monitored in this investigation were sedentary for a substantial portion of each day.

It is challenging to accurately account for daily energy expenditure in free-living organisms. The daily energy requirement for weight maintenance of a healthy adult animal includes metabolic demands for basic life processes (typically estimated as resting energy expenditure) and energy expended for thermoregulation, assimilation of nutrients from the diet, and physical activity. The first 3 of these components can be measured accurately in controlled conditions. However, it is more difficult to measure the last component (energy expended for physical activity), particularly in animals in a free-living situation. Methods that involve the use of indirect calorimetry are not feasible in nonhuman subjects (measurement of gas-exchange by use of portable metabolic systems) or are cost-prohibitive for use on a large scale (doubly labeled water). The most common method used to determine energy requirements is to establish the amount of a food of known energy content that is required to maintain energy balance in a subject. Although this is an approach that could potentially be applied rigorously to companion animals in a home setting, these types of determinations are, with few exceptions, made in kenneled animals.

Consequently, feeding recommendations found on pet food labels or those provided by veterinary health professionals are determined primarily on the basis of predictive equations, and a common approach is to use a factorial method that involves a calculation of resting energy expenditure, which is then multiplied by factors that allow for physiologic state, amount of activity, and other variables that could affect total daily energy expenditure. Voluntary physical activity can vary considerably among individual animals; as such, it is likely the aspect of the factorial method that involves the most speculation when estimating the energy requirements of pet dogs.

Various approaches can be used to quantify physical activity. The simplest approach is for an owner to maintain a record of the amount of time a pet spends engaged in various types of activities throughout a day. This method has been used in several investigations; however, it is cumbersome, time-consuming, and unlikely to yield an accurate account of total activity in a 24-hour period.5–7

In the past few decades, technological advances in motion sensing have led to the development of small, unobtrusive devices that can detect and record physical activity for extended periods. There are basically 2 classes of activity monitors: those that detect steps (pedometers) and those that detect changes in acceleration (accelerometers). Modern electronic pedometers are extremely accurate for recording the number of steps and distance traveled in human subjects.8 A study9 has been published on the use of pedometers as a tool for measuring physical activity in dogs. In that study,9 the investigators found that for standardized conditions, there was no significant difference between the number of steps recorded by the pedometer and the actual number of steps taken (as determined by videographic analysis) when dogs trotted or ran; however, a significant difference was detected during walking. Furthermore, the correlation between the number of steps recorded by the pedometer and the owner-reported amount of time spent in purposeful activity was poor (r = 0.305). Further investigation of the use of pedometers in dogs may be able to address these issues; however, these devices have some inherent limitations, including the inability to provide information on the duration, frequency, and intensity of physical activity. Accelerometers are capable of detecting those aspects of physical activity; consequently, these devices have been widely investigated in human subjects8 and, more recently, in dogs.1,3

Results of the study reported here revealed that an accelerometer-based activity monitor can be used to discriminate effectively among standardized activities in pet dogs; however, only sedentary activity (remaining in sternal or lateral recumbency for the entire data collection period) can be equated to reflect spontaneous activity of similar intensity in free-living dogs. It will be necessary to compare counts from the standardized activities in phase 1 with activities of light and moderate to vigorous intensity in free-living dogs to determine validity of our cut points. It should also be recognized that 1 inherent limitation in the use of accelerometers as an indirect means of measuring energy expended on physical activity is that they do not detect the energy expended against gravitational forces (eg, energy expended to stand and remain standing, to move up an incline, or to pull a load).

Thus, until the accelerometer used in the present study has been validated for measurement of activities in dogs in a free-living situation to verify the accuracy of the cut points for light and moderate to vigorous activities, our findings must be considered speculative. This is further underscored by the fact that the scope of activities in which dogs may engage (eg, jumping, running, or roughhousing) was not necessarily reflected by the controlled activities we used as standards for light and moderate to vigorous activities. However, the cut point for sedentary activity should be valid, and consequently, our data suggest that the pet dogs monitored for 14 days in this investigation were sedentary for a substantial portion of each day and that the amount of time that dogs are sedentary increases as they age. There are 2 caveats regarding this statement. First, because we set the epoch length at 1 minute for data collection during the 14-day period, it is possible that activities of greater intensity that dogs engaged in only briefly were not always detected. The second caveat is that dogs participating in the study were principally owned by staff and students of the School of Veterinary Medicine at the University of Pennsylvania; therefore, the amount of activity for these dogs may not reflect that of dogs living elsewhere in the United States or belonging to people with differing lifestyles.

With regard to the application of activity monitoring to the factorial method of estimating energy require ments in dogs, detailed food diaries were maintained for the dogs that participated in phase 2 of the study for the same 14-day period during which they wore the activity monitor. Analysis of those data is ongoing, but those results may help determine whether the percentage of time a dog spends in activities of differing intensity is correlated with the caloric intake required to maintain it in energy balance. Increasing the accuracy for application of activity factors in the calculation of a pet dog's daily energy requirement will lead to better feeding recommendations, whether those recommendations are on pet food labels or provided by veterinary health professionals. More accurate feeding recommendations should be of substantial benefit to pet dogs in light of the prevalence of obesity and its health consequences in this population.10

In the study reported here, we concluded that activity counts recorded by an accelerometer-based activity monitor can be used to discriminate effectively among standardized activities in pet dogs. Further investigations will be necessary to verify whether the cut points for light and moderate to vigorous activities that were established from the standardized activities in phase 1 are valid for detecting activity of similar intensities in free-living dogs.

ABBREVIATIONS

AUC

Area under the curve

ROC

Receiver operating characteristic

a.

Actical, Mini Mitter Inc, Bend, Ore.

b.

Stata, version 8, StataCorp, College Station, Tex.

References

  • 1.

    Brown DC, Michel KE, Love M, et al. Evaluations of the effect of signalment and body conformation on activity monitoring in companion dogs. Am J Vet Res 2010; 71: 322325.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 2.

    Dow C, Michel KE, Love M, et al. Evaluation of optimal sampling interval for activity monitoring in companion dogs. Am J Vet Res 2009; 70: 444448.

  • 3.

    Hansen BD, Lascelles BDX, Keene BW, et al. Evaluation of an accelerometer for at-home monitoring of spontaneous activity in dogs. Am J Vet Res 2007; 68: 468475.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 4.

    Greiner M, Pfeiffer D, Smith R. Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests. Prev Vet Med 2000; 45: 2341.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 5.

    Slater MR, Robinson LE, Zoran DL, et al. Diet and exercise patterns in pet dogs. J Am Vet Med Assoc 1995; 207: 186190.

  • 6.

    Bauman AE, Russell SJ, Furber SE, et al. The epidemiology of dog walking: an unmet need for human and canine health. Med J Aust 2001; 175: 632634.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7.

    Robertson ID. The association of exercise, diet and other factors with owner-perceived obesity in privately owned dogs from metropolitan Perth. Prev Vet Med 2003; 58: 7583.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 8.

    Bassett DR. Validity and reliability issues in objective monitoring of physical activity. Res Q Exerc Sport 2000; 71: 3036.

  • 9.

    Chan CB, Spierenburg M, Ihle SL, et al. Use of pedometers to measure physical activity in dogs. J Am Vet Med Assoc 2005; 226: 20102015.

  • 10.

    Lund EM, Armstrong PJ, Kirk CA, et al. Health status and population characteristics of dogs and cats examined at private veterinary practices in the United States. J Am Vet Med Assoc 1999; 214: 13361341.

    • Search Google Scholar
    • Export Citation
All Time Past Year Past 30 Days
Abstract Views 146 0 0
Full Text Views 1610 774 77
PDF Downloads 539 161 11
Advertisement