Long-term measurement of cardiovascular variables by telemetry in laboratory animals has become indispensable in recent years. However, limited battery life and management of large volumes of recorded data are major drawbacks. These limitations can often be overcome by intermittent sampling of data. The question is, how much data does one need to collect to accurately reflect the underlying average value? To investigate this, 24-h continuous recordings of rabbit heart rate, arterial pressure, and integrated renal sympathetic nerve activity (RSNA) were resampled using a variety of protocols that differed with respect to the number of individual sampling periods used and the total amount of time that was sampled. The absolute percentage errors of estimates of the daily mean, standard deviation, and interquartile range were calculated for each sampling protocol. A similar analysis was repeated using arterial pressure data from rats. The results show that the number of sampling periods spread throughout the day had more effect than the total amount of data recorded. For example, just 2 h of total sampling time spread over 12 evenly spaced 10-min periods gave estimates of the daily mean of blood pressure and heart rate with < 1% error and RSNA with < 3% error. We show that accurate estimates of the daily mean of arterial pressure, heart rate, and RSNA can all be made using scheduled recording, and we recommend recording a minimum of 2 h/day spread over a number of periods throughout the day.
- blood pressure
- sympathetic nerve activity
the use of telemetry to assess the heart rate and blood pressure of laboratory animals has become an indispensable part of cardiovascular physiology, drug development, and safety pharmacology (9). The ability to record cardiovascular signals in freely moving animals in their home cages over extended periods of time has opened up new areas of research. Telemetry devices to record blood pressure and heart rate have been available for some time (3), and recently new telemetric devices have been developed that allow both sympathetic nerve activity (SNA) and blood pressure to be measured in a variety of animal models (2). The development of such telemetry-based devices has revolutionized the recording of physiological signals by removing the need for stress-inducing tether arrangements and reducing potential sources of infection associated with external catheters or electrode leads. However, the length of recordings that can be made with such devices can be limited by battery life. While improving all the time, the power consumption of the high bandwidth transmitters required for high-frequency signals such as SNA exceeds the limits of current capacity of small batteries if recordings are to be made for significant periods (e.g., > 1 month). Other methods of supplying power, such as using wireless power (4, 12) are being developed for use in rodents, such as the rat, but for a number of reasons their application to larger animals, such as the rabbit, is limited. Intermittent sampling, with the telemeter turned off between sample periods, is obviously one approach for spreading limited battery capacity over extended recording periods.
A second difficulty in long-term monitoring of physiological signals by telemetry (or other methods) is the need to store and process large volumes of data obtained. For example, sampling the blood pressure waveform at 200 Hz for 3 days leads to data files in excess of 200 Mb. For SNA, a signal thought to contain information up to 5,000 Hz, continuous sampling quickly produces unwieldy amounts of raw data. Again, intermittent sampling of data is a possible solution being frequently used to reduce the data collected over long recording periods.
While intermittent sampling may greatly facilitate long-term recordings of cardiovascular variables by telemetry, a crucial question is, what is the minimum amount of sampling of cardiovascular variables required to get an accurate estimate of their long-term mean value? In the present study, we assessed the impact of different sampling protocols on the accuracy of estimates of the 24-h mean level of blood pressure, heart rate, and SNA recorded by telemetry in rabbits. In addition, we repeated the main part of our analysis using rat blood pressure data. We show that accurate estimates of the daily mean of arterial pressure, heart rate, and renal SNA (RSNA) can all be made using scheduled recording.
Experiments were conducted in six New Zealand White rabbits with initial weights of 2.4 to 3.5 kg and were approved by the University of Auckland animal ethics committee. Rabbits were housed individually in cages (height 45 cm, width 65 cm, and depth 65 cm) with a telemetry receiver (model RLA2000; Data Sciences International, St Paul, MN) positioned on the side of each cage. The rabbits were fed daily (100 g standard rabbit pellets, supplemented with hay, carrot, and apple) at 0900, and water was available ad libitum. The room was kept at a constant temperature (18°C) and light/dark cycle (lights on from 0600 to 1800).
Full experimental results have been reported previously along with surgical details (2). Using halothane anesthesia, rabbits were instrumented with a radiotelemetry transmitter to record arterial pressure (model PA-D70; Data Sciences International). The telemeter catheter was inserted into a branch of the left iliac artery and advanced so that its tip lay in the abdominal aorta 1 cm above the iliac bifurcation but well below the renal artery. During a second surgery, 2 wk after the first, a telemetry-based implantable nerve amplifier (model 2003/01; Telemetry Research, Auckland, New Zealand) was inserted via a flank incision with the electrodes coiled around the left renal nerve and the electrode and nerve coated in a silicone elastomer (Kwik-sil; World Precision Instruments, Sarasota, FL) (2).
The rabbits were allowed to recover from surgery to implant the nerve electrodes for 1 wk before data collection began. The recording of arterial pressure and RSNA via telemetry allowed monitoring to take place remotely with rabbits housed in their home cages. RSNA signals were amplified between 10 and 20,000 times and band-pass filtered between 50 and 5,000 Hz; this signal was used for audibly checking the quality of the recording. The amplified signal was also full-wave rectified and integrated with a time constant of 20 ms. Subsequent analysis was performed on this integrated signal, presented as arbitrary units (au). All data were sampled at 500 Hz using an analog-to-digital data acquisition card (model AT-MIO64E-3; National Instruments, Austin, Texas). All subsequent data collection and analyses were performed using a data acquisition program (Universal acquisition and analysis version 11; Telemetry Research) as described previously (2). Signals were initially sampled at 500 Hz to derive 2-s averages of heart rate, mean blood pressure, and RSNA. These 2-s averages were used for all subsequent analysis.
Three separate days of data from the control periods of each of the six rabbits were used. Each file contained a 24-h (0000–2400) recording of heart rate, arterial pressure, and integrated RSNA consisting of 43,200 sequential 2-s averages. The complete 24-h data set was used to calculate the daily mean, standard deviation (SD), and interquartile range (75–25th percentiles) of each variable. Then, using MatLab (The Mathworks, Natick, MA), each day's data were resampled using one of a number of sampling protocols, and this resampled data was used to produce estimates of the daily mean, SD, and interquartile range. The absolute error associated with a given sampling protocol was determined as the absolute difference between the estimate obtained using the resampling protocol data and the mean value calculated using the complete 24-h data set and was expressed as a percentage of this latter value (Fig. 1). For each rabbit, the absolute percentage error was calculated for each of the 3 days of data that were then averaged to give a mean error for each animal. The mean ± SE was then calculated for the entire group.
Sampling protocols were defined by the total period of sampling duration (12, 8, 6, 4, 3, 2, 1, 0.2 h) and the number of “on periods” that these data were spread throughout the day (1–120/day). For example, 12 h of data can either be collected as a single 12-h recording period or spread throughout the day as 12 1-h recording periods (Fig. 2). The use of 10-s recording periods was also investigated by calculating the error that would occur if 10 s of data were recorded at intervals ranging from every 30 s to once every 24 h. The first “on period” for each sampling protocol started at midnight each day.
Arterial pressure recording from a subgroup of rats.
A similar analysis was performed using blood pressure data obtained from a control group of six Long-Evans rats during the course of a previous study (22). Briefly, the rats were instrumented with radiotelemetry transmitters to record arterial pressure (model TA11PA-C40; Data Sciences International), which was sampled in 3-s bursts once every 10 s (i.e.. 7,920 per day). The daily mean arterial pressure level for each rat was calculated using all 7,920 blood pressure values recorded during the 22-h period. The data were then resampled to investigate the accuracy with which the 22-h mean could be estimated using data collected at intervals ranging from 1 per minute to once every 8 h. The absolute percentage error was then determined as the absolute difference between an estimated daily values and the daily mean value calculated using the complete data set sampled at 10-s intervals.
Effect of sampling on the accuracy of estimates of the daily average level of cardiovascular variables in rabbits.
The average arterial pressure, heart rate, and RSNA values for the group were 82 ± 5 mmHg, 241 ± 8 beats/min, and 27 ± 5 au, respectively. The error in estimating the 24-h mean of arterial pressure varied among individual sampling protocols, ranging from 4.5 ± 0.4% for a single 12-min sampling period to 0.06 ± 0.01% for 120 6-min sample periods (total recording time = 12 h) (Fig. 3A). Using the same sampling protocols, the error for estimates of the 24-h mean ranged between 7.5 ± 1.6% and 0.10 ± 0.02% for heart rate (Fig. 3B) and 13.2 ± 4.6% to 0.5 ± 0.1% for RSNA (Fig. 3C). For each of the three variables, the trend was for the error in estimating the 24-h mean to fall as the total amount of time spent sampling increased. This trend was particularly pronounced for total sampling times < 2 h/day (Fig. 3). For any given amount of total sampling time, the trend was for the error to fall as the sampling time was divided up into a greater number of individual periods distributed over the day.
Effect of sampling on the accuracy of estimates of the daily variation of cardiovascular variables in rabbits.
The average SD of blood pressure, heart rate, and RSNA values for the group were 7 ± 1 mmHg, 28 ± 2 beats/min, and 12 ± 3 au, respectively. As was the case for estimates of the daily average level of cardiovascular variables, estimates of daily variation fell with increasing total sampling time and with the number of individual sampling periods over the course of the day. The absolute percentage error for estimates of the SD of blood pressure ranged from 36 ± 6% for a single 12-min period to 1.3 ± 0.3% for 120 6-min periods (total recording time = 12 h) (Fig. 4A). Using the same sampling protocols, the error for heart rate ranged between 36 ± 3% and 0.9 ± 0.2% (Fig. 4B), and for RSNA the range was 22 ± 4% to 2 ± 0.6% (Fig. 4C). In general, the level of error associated with estimates of the SD of each of the three variables was considerably greater than that associated with the corresponding estimates of the daily mean.
A highly similar pattern was observed for estimates of the interquartile range. The interquartile range of blood pressure, heart rate, and RSNA over the 24-h period averaged 10 ± 1 mmHg, 41 ± 3 beats/min, and 13 ± 3 au, respectively. In the case of blood pressure, the absolute percentage error of estimates of the 24-h interquartile range amounted to 45 ± 6% for a single 12-min period and 1.9 ± 0.3% for 120 6-min periods (total recording time = 12 h). Using the same sampling protocols, the error for estimates of the interquartile range of heart rate amounted to 50 ± 4% and 1.2 ± 0.3%, and 24 ± 5% to 1.3 ± 0.5% for RSNA.
Effect of the number of 10-s sampling periods on estimates of the daily average level of cardiovascular variables in rabbits.
Intermittently repeated 10-s sampling periods have often been used in cardiovascular monitoring. In the case of RSNA, the error associated with estimates of the 24-h mean level fell considerably as the number of sample periods per day was increased, ranging from 34.9 ± 17.0% for one 10-s sample per 24 h to 0.18 ± 0.06% for one 10-s sample every 2 min (720/day) (Fig. 5). The absolute percentage error associated with estimates of the 24-h mean of arterial pressure and heart rate were consistently lower but fell in the same manner as the number of sample periods per day was increased, ranging from 8.3 ± 1.4% for one 10-s sample per 24 h to 0.03 ± 0.01% for one 10-s sample every 2 min and 11.3 ± 1.7% for one 10-s sample per 24 h to 0.03 ± 0.004% for one 10-s sample every 2 min (Fig. 5).
Effect of the number of 3-s sampling periods on estimates of the daily average level of blood pressure in rats.
Arterial pressure data from a group of six rats showed a similar pattern to the rabbit data. Fig. 6 shows the absolute percentage error in calculating the 22-h mean of arterial pressure As was the case with rabbit data (Fig. 5), absolute error decreased as the number of recording periods and total recording time increased (Fig. 6).
Long-term continuous monitoring of cardiovascular variables provides a wealth of information but leads to unwieldy amounts of data and, in the case of some telemetry applications, may rapidly exhaust limited battery capacity. In this study we set out to investigate the impact of scheduled intermittent sampling protocols and their ability to accurately estimate the most commonly used statistical descriptors of cardiovascular variables (i.e., the mean, SD, and interquartile range). Using blood pressure, heart rate, and RSNA data previously collected from the rabbit as continuous 2-s averages, we applied a variety of sampling protocols to determine the effect of both the total amount of data collected and the distribution of this data throughout each 24-h period on the accuracy of estimates of the daily mean, SD, and interquartile range. Not surprisingly, increases in both the total recording time and the frequency of individual sampling periods were associated with more accurate estimates of the daily mean values.
One important finding from our analysis of the rabbit data is that it indicates that there is little benefit to recording more than a total of 2 h per day (Figs. 3 and 4). Instead, our data suggest that increasing the number of periods over which the recording is distributed will have a greater effect on the accuracy of the estimates of the daily mean and the SD values than increasing the total amount of recording time beyond 2 h per day.
It has long been appreciated that spreading recordings of blood pressure and related variables over the course of the day will better capture circadian and other variations and therefore provide a more representative estimate of the daily mean value (6, 7, 14, 17, 20). This is clearly portrayed in Fig. 2 where the same amount of data is plotted either as a single 6-h period or 30 min every 2 h. The influence of the circadian rhythm on all three variables becomes clear from the shape of the graphs in Figs. 3 and 4. These results show that spreading the data recording throughout the day will have a greater effect on reducing the error than simply recording longer over a single period. For example, just 12 min of recording time spread over 120 periods/day gave a significantly more accurate estimate of the daily mean of blood pressure (error = 0.5 ± 0.1%) than recording over a single 12-h period (error = 1.9 ± 0.4%).
In addition to the errors associated with inadequate sampling, measurements made over brief periods or a single point in time often also suffer inaccuracies as a result of stress associated with disturbance or handling of the animal (9). However noninvasive the measurement appears to be, if the animal must be handled, there will inevitably be an element of stress that may have significant effects on the variables recorded (8, 9, 13). Studies have shown that hypertensive animals may have an augmented response to these stresses (8) making comparison between groups difficult. Tail cuff measurements in sinoaortic denervated rats initially lead researchers to believe that these animals are hypertensive [mean arterial pressure (MAP) = 156 ± 5.4 mmHg], whereas continuous measurements via telemetry have shown that they are in fact normotensive (MAP = 119 ± 4.7 mmHg) but have elevated arterial pressure during restraint (15). In this case, the tail cuff measurement overestimated the MAP by ∼31%. There is little doubt that using telemetry is an attractive alternative to such methods, but until now researchers have been uncertain of how much and how often to record data to accurately reflect the underlying daily mean. The results from this study show that even if telemetry is used to collect data for only a short period each day, the results can be expected to be more accurate than those collected by tail cuff measurement.
The results of this study show that the mean levels of all three variables, but particularly arterial pressure, is well represented by surprisingly few data points. For example, using one 12-min sampling period resulted in an error in mean arterial pressure of only 4.5 ± 0.4%. However, the error in calculating the estimates of the daily mean of heart rate and SNA were larger than those of blood pressure. This is not unexpected given that the variability of the SNA signal (SD = 44% of mean, SD 12 ± 3 au, mean 27 ± 5 au) is the greatest, followed by that of heart rate (SD = 12% of mean, SD 28 ± 2 beats/min, mean 241 ± 8 beats/min) and then blood pressure (SD = 8% of mean, SD 7 ± 1 mmHg, mean 82 ± 5 mmHg).
One caveat to the present analysis is that, while very short regular sampling of data will reflect the underlying mean, other forms of analysis, such as spectral analysis of low-frequency oscillations, heart rate variability analysis, estimates of spontaneous baroreflex activity, and transfer function estimates of baroreflex gain often require much longer periods of continuous data to be collected. The dynamic properties of blood pressure and other cardiovascular variables are often used to help investigate cardiovascular control mechanisms. For example, we have used spectral analysis of 5-min blood pressure and RSNA recordings in intact and arterial-baroreceptor denervated rabbits to investigate the role of the arterial baroreceptors in the timing of SNA bursts and their entrainment to the blood pressure pulse wave (11). Using very short bursts of recording (e.g., 30-s periods) may not allow for the necessary information to be gathered from the signal. For some, the accurate estimates of the SD over a 24-h period that can be achieved through frequent use of short bursts of recording (Fig. 4) will be sufficient, but for others it may not. In practice, for most applications a balance will need to be met between spreading recordings in shorter periods throughout the day to reduce the error in calculating the 24-h mean and satisfying requirements for other types of analysis. For example, recording for 15 min every 2 h would provide high accuracy while maintaining a reasonable length of individual sample periods (15 min) for additional analyses.
For the purposes of this study, no consideration was given to the time of day that recordings were made. Each day's data file began at midnight, and sampling protocols all began at this time. This meant that where a single recording period was used the data were taken from the period beginning at midnight in all cases. However, in practice, the user may wish to consider the timing of the recording periods more closely. For example, some researchers use a 20-h average as the daily mean to allow for the period of the day where the animals may be disturbed by animal husbandry tasks, such as feeding and cage cleaning (e.g., Refs. 5, 14). It may be prudent to consider these periods of disturbance and to schedule recording sessions to fall outside of these times where possible.
While most of the data of this study were collected from the rabbit, the results should be broadly applicable to any species. For rodents, like the rat, it would be expected that spreading the recording periods throughout the day may have an even larger effect on the error in the estimates of the daily means due to the larger circadian rhythm seen in the heart rate of these animals compared with the rabbit (16). To investigate the effect of scheduled recording on rat data we applied similar analysis to data previously collected (22). The plot in Fig. 6 shows a similar pattern to the rabbit data with increasing frequency of the recording bursts giving a reduction in the calculated error.
When considering the application of these results to species other than rabbits and rats the frequency content of the signals in question, including the heart rate of the animal, should also be taken into account with respect to the sampling frequency used and the way data is collected. The rabbit data used in this study were collected at 500 Hz with the data then saved as 2-s averages. The use of 500 Hz will easily capture the arterial pressure waveform of most species to allow accurate timing of the heart beat and provide good estimates of mean arterial, systolic, and diastolic pressures and heart rate. Even the blood pressure and heart rate of the mouse, with a heart rate of up to 12 beats/s, is well captured by 500 Hz sampling (23). However, while averaging over a 2-s period in the mouse will capture ∼24 heart beats and in a rabbit ∼8 heart beats, animals with slower heart rates may require longer periods. For example, in dogs with a heart rate of ∼90 beats/min (6) a 2-s average will only capture three heart beats, so it may be more appropriate to use a longer period e.g., average the 500 Hz sampled data over 4 s.
One of the commonly used telemetry blood pressure systems in physiological research is the Physiotel system by Data Sciences International. This was one of the first blood pressure telemetry systems available and is widely used in a number of species including the rabbit (e.g., Refs. 1, 5, 18), rat (e.g., Refs. 3, 19, 21), and mouse (e.g., Refs. 10, 20, 23). The Dataquest software supplied by the company allows one to either record continuously or as short bursts (typically 10 s) from each animal in turn. This is then repeated at set intervals throughout day. However, as with other systems that enable chronic blood pressure measurement through swivels, for example, there has been little guidance for the end user on how to select total recording time, burst frequency, and burst time to render scheduled sampling a dependable substitute for continuous measurement. From the results in Fig. 5, we can confidently suggest that if data are recorded for 10 s every 15 min, the daily means of blood pressure, heart rate, and RSNA of the rabbit will be well estimated with errors of 0.4 ± 0.1%, 1 ± 0.1%, and 2 ± 0.4%, respectively. This is a good example of how scheduled sampling can allow recording from more animals using fewer resources.
With telemetry being used for chronic measurement of cardiovascular variables over increasingly longer periods, the management of both the data collected and battery life are becoming a significant issue. In this study, we have shown that accurate estimates of the daily mean of arterial pressure, heart rate, and RSNA can all be made using scheduled recording. Our recommendation is to record for a minimum of 2 h per day and to spread this recording over a number of periods throughout the day.
This work was supported by the Auckland Medical Research Foundation and the Health Research Council of New Zealand.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2008 the American Physiological Society