Address for correspondence Anneke M.P. De Schryver MD, Department of Gastroenterology, University Medical Center, F02.618, PO Box 85500, 3508 GA Utrecht, The Netherlands. Tel.: + 31 30 250 7004; fax: + 31 30 250 5533; e-mail: email@example.com
AbstractThe aims of this study were to explore all characteristics of high-amplitude propagated contractions (HAPCs) that would allow them to be distinguished from nonHAPC colonic pressure waves, and to develop computer algorithms for automated HAPC detection. Colonic manometry recordings obtained from 24 healthy volunteers were used. Automated analysis was performed to detect propagated pressure waves and to determine their amplitude, duration and area under the curve (AUC). For each of these variables distribution plots were made. Automated HAPC counts were compared to visual counts by experienced investigators. Distribution plots of 141093 colonic pressure waves lacked a bimodal pattern, as was also the case for propagated contractions (n = 8758). With increasing high-amplitude thresholds for HAPC detection, a gradual decrease in the automatically detected HAPC number was observed. These findings precluded determination of a threshold. Taking visually detected HAPCs as reference, amplitude thresholds of 100 mmHg in two channels, and 80 mmHg in one channel yielded the highest sensitivity (92%). In conclusion, objective criteria to distinguish HAPCs from other propagated pressure waves on the basis of their amplitude, duration or AUC do not exist. Automated detection of HAPCs using empirically derived criteria leads to an acceptable degree of correlation with visually detected HAPCs.
Prolonged colonic manometry is a specialized technique with which colonic motility can be studied in humans. Some characteristics of colonic motility can easily be distinguished by visual inspection such as a decreased motor activity during the night, and the presence of high-amplitude propagated contractions (HAPCs), which occur mainly after waking up but also later during the day.
Several studies on human colonic motility have shown that HAPCs are associated with defecation.1,2 The incidence of HAPCs in patients with constipation is lower than in healthy volunteers (2.6 ± 0.7 day−1 vs. 6.1 ± 0.9 day−1), suggesting a relation between the incidence of bowel movements and that of HAPCs.1,3 However, colonic propulsive activity is probably due to a spectrum of propagated waves, as recently shown by a study taking into account low-amplitude propagated contractions.4 In another recent study, Cook et al.5 used a scintigraphic technique to demonstrate that transport of bowel contents is related not only to propagating but also to nonpropagating pressure wave sequences. The former, however, accounted for a greater extent of isotope movement.
Although most investigators in this area would agree that HAPCs are clearly identifiable manometrically by their high amplitude, long duration and propagation over several channels, different groups of investigators use different criteria to identify HAPCs (Table 1). In the published studies, amplitude thresholds ranging from 50 to 135 mmHg have been used.6–11 In most studies, a rationale for the selected amplitude threshold is not given, with two exceptions. Cook et al.9,12 and Rao et al.10 calculated the mean amplitude + 2SD of the amplitude values of all propagated pressurewaves and used this value (90 and 105 mmHg, respectively) as a cutoff.9 Because propagated contractions with amplitudes lower than those of HAPCs can also propel luminal contents, the propulsive characteristics of pressure waves cannot be used to distinguish HAPCs.5,6
Table 1. HAPC definitions used in published colonic manometry studies
Minimum no. of channels
Distance between channels (cm)
Minimum duration (s)
Velocity (cm s−1)
*xx, Propagating over most of the length of the colon.
In the present study we aimed to enlarge our understanding of HAPCs by investigating phasic motility in the left colon during approximately 40 h in 24 healthy volunteers using solid-state colonic manometry. It was hypothesized that HAPCs have features that are different from the remaining colonic propagating pressure waves and thus are not just part of a continuum of propagated contractions.
The main objective of this study was to explore all features of HAPCs that would possibly allow their distinction from other colonic pressure waves. Furthermore, we aimed to develop computer algorithms to detect HAPCs in an automated fashion.
Materials and methods
In this study, colonic manometry data that had beenobtained previously from 24 healthy volunteers (11males) ranging in age from 26 to 72 years (mean age 45 years) were used. All volunteers were judged to be healthy following physical examination prior to entry in the study. None of them had a history of gastrointestinal symptoms, use of medications, or previous abdominal surgery. Criteria for participation included a normal bowel habit, i.e. a bowel frequency of between three stools daily and three stools weekly, with a normal consistency and without straining. All subjects were informed of the nature, general purpose, and possible risks involved in the study before giving their written informed consent to participate. The ethics committee of the University Medical Center, Utrecht had approved the study protocols.
The colon was cleansed using 2 L of soap-water enema, administered 3 h before the procedure.
After placement of the catheter the recording started. The manometric recordings were carried out over a 40-h period, starting on day 1 at approximately 16.00 h and ending on day 3, at about 08.00 h.
During the day, subjects were fully ambulant. Standard meals were provided for the recording period. These meals consisted of a dinner (630 kcal) on day 1, breakfast (529 kcal), lunch (707 kcal) and dinner (731 kcal) on day 2 and a breakfast (529 kcal) on day 3. The subjects went to bed at 23.00 h and woke up at 07.00 h. Only the 24-h period from the second (06.00 h) until the third day (06.00 h) was used for analysis.
Colonic motility recordings were performed using a catheter (total length 230 cm, outer diameter 3.3 mm) with six solid-state pressure transducers, located at 2, 12, 22, 32, 42 and 52 cm from the tip (Millar Instruments, Inc. Houston, Texas, USA). Pressure transducers were calibrated at pressures of 0 and 60 cm H2O, using a vertical cylinder filled with water.
The catheter was introduced colonoscopically, without premedication, by advancing it together with theendoscope. The tip of the probe was fixed to the tip of the colonoscope by a thread held by a loop maintained inside the operative channel of the endoscope. Once the transverse colon was reached (checked fluoroscopically) the tip of the catheter was released and the colonoscope was gently withdrawn, leaving the catheter in situ. During the withdrawal, air was aspirated as completely as possible. At the end of the colonoscopy the catheter position was checked by fluoroscopy. This was done again on day 2 and on day 3, before removing the catheter.
After placement, the catheter was taped firmly to the back of the subject, and connected to a portable datalogger (Medical Measurement Systems, Enschede, The Netherlands). The storage capacity of the datalogger is 4 MB, the sample frequency was 4 Hz. The datalogger was carried in a shoulder bag, allowing continuous recording with a freely moving subject. Data were transferred to an IBM-compatible computer for further display and off-line analysis, using commercially available software (Medical Measurement Systems, Enschede, The Netherlands) and software developed by the investigators.
Artefacts caused by intra-abdominal pressure changes (characterized by simultaneous pressure peaks in all six channels) were eliminated using previously described technique of subtraction of a so-called minimum curve.13 Colonic baseline values were calculated over 1-min stretches but when the difference in baseline value between two consecutive 1-min periods was larger than 1 kPa, subsequent baseline calculation was performed using 10-s periods. Peak detection was carried out using the baseline curves as a reference. Pressure waves were detected and included in analysis when they had a minimum duration of 3 s and reached a threshold of 1.3 kPa (= 10 mmHg). Furthermore, a propagation analysis was performed, leading to detection of all pressure waves (defined as pressure waves of > 10 mmHg that propagated over at least three (consecutive) channels, with a velocity of > 0.2 cm s−1 and < 10.0 cm s−1).
In all manometric recordings (day 2, 06.00 h to day 3, 06.00 h) the detected colonic pressure waves were analysed for amplitude, duration and area under the curve (AUC). For each of these variables, distribution plots were created in search of a bimodal pattern that would be suggestive of the existence of two populations of contractions: HAPCs and the remainder of pressure waves. Distribution plots were also made of all detected propagated pressure waves, for the three variables mentioned above (amplitude, duration and AUC). Finally, the means + 2SD and 95th percentiles for amplitude of all pressure waves and of the propagated pressure waves were determined.
Visual analysis of HAPCs
In order to facilitate the validation of HAPC analysis, from each of the recordings 20-min segments that contained pressure waves with an amplitude > 6 kPa (= 45 mmHg) were selected. These were spliced electronically to form one new data file. The threshold of 6 kPa was chosen because 50 mmHg (= 6.7 kPa) is the lowest amplitude threshold for HAPC detection described in literature. In total, 108 20-min segments of colonic manometric recordings were compiled to a 36-h datafile. This 36-h compilation file was analysed manually by three experienced investigators, without making use ofpredefined criteria. The observers were requested to detect ‘clearly identifiable large waves that were propagated over three or more channels’. They were blinded to the results of automated HAPC analysis. HAPCs detected by all three observers and by two of the three observers were identified.
Automated analysis of HAPCs
Automated analysis of the 36-h compilation datafile was performed by software developed in our department. The propagation analysis algorithm searched forpropagated peaks (propagating over at least threechannels, with a velocity of > 0.2 cm s−1 and < 10.0 cm s−1) with an amplitude above a specified threshold. The algorithm first identified the propagated waves with the highest amplitude within the given time window. Subsequently the algorithm searched forprogressively smaller propagated peaks in adjacent channels.
Distribution of peak amplitude, duration and AUC
In the 24-h 6-channel datasets studied, a total of 141093 colonic pressure waves was detected. As shown in Fig. 1 (left panel), the distribution plots for amplitude, duration and AUC all showed a skewed distribution but failed to show any sign of a bimodal pattern. This was also the case for distribution plots of individual patients' data (not shown).
Propagation analysis yielded 8758 sequences of pressure waves that were propagated over at least three channels. Distribution plots of amplitude, duration and AUC of these propagated pressure waves did not show a bimodal pattern either, again suggesting that HAPCs lack characteristic features (Fig. 1, right panel).
Visual analysis of HAPCs
Visual analysis of the 36- h colonic manometry compilation by three experienced investigators yielded a total number of 192, 170 and 171 HAPCs, respectively (mean 178 ± 7.2). Two of the three observers detected 175 HAPCs, whereas a total number of 152 HAPCs was identified by all three observers.
Automated analysis of HAPCs
An example of the automatic detection of HAPCs is shown in Fig. 2. As expected, the number of HAPCs detected was dependent on the amplitude threshold used in the detection algorithm. When the amplitude threshold was increased stepwise from 45–150 mmHg, the number of detected HAPCs decreased from 386 to 70 (Fig. 3). The curve of HAPC numbers as a function of detection threshold showed a gradual decline, which again provided no clue as to the optimal detection threshold.
For the automated HAPC detection to yield numbers of HAPCs that corresponded to the mean number of HAPCs detected by the three observers (n = 178) and to the total numbers observed by two and three observers (n = 175, n = 152) the amplitude thresholds had to be set at 85, 85 and 100 mmHg, respectively (Fig. 3). The automated analyses yielded sensitivities of 85.4%, 86.9% and 94.1%, respectively (using visual analysis as gold standard), whereas the specificity was > 98.5% in all three cases (98.6%, 98.7% and 99.5%, respectively).
Although the sensitivity of automated analysis was highest (94.1%) when the set of HAPCs detected by all human observers was used as the reference, the number of thus detected HAPCs (n = 152) was far below the mean number of visually detected HAPCs (n = 178). The discrepancy between the automated and visual recognition of HAPCs appeared to be largely due to HAPCs that had high-amplitude pressure components in two channels but a subthreshold pressure wave in the third channel. When the amplitude threshold in one channel (regardless which one) was lowered to 90, 80 and 70 mmHg, 166, 179 and 192 HAPCs were detected, with accompanying sensitivities of 90.4%, 91.9% and 88.0%, respectively. Thus, the optimal performance of automated HAPC detection, regarding both the total HAPC count and sensitivity, was achieved by detecting pressure waves that propagated over at least three channels with minimum amplitudes of 100 mmHg in two channels and 80 mmHg in one channel. This detection had a specificity of 99.1%. With these amplitude thresholds, the number of automatically detected HAPCs per subject was 7.5 ± 1.4 (mean ± SEM) per 24 h. The amplitude of these HAPCs was 179 ± 4 mmHg, their duration was 16.5 ± 1.2 s and their propagation length was 36.3 ± 0.8 cm. The propagation velocity of HAPCs was 1.25 ± 0.10 cm s−1 (range 0.39–5.71).
The 95th percentile for amplitude for all automatically detected pressure waves was 61 mmHg, whereas the 95th percentile for amplitude of only propagated pressure waves was 55 mmHg. Calculation of the mean + 2 SD for the amplitude of all automatically detected propagating pressure waves yielded an amplitude of 47 mmHg.
Although HAPCs are said to be easily recognizable contractions, the many different definitions used to identify them indicate that there is no consensus about the criteria for HAPC detection.
This is the first study that shows that HAPCs cannotbe distinguished from other propagated colonic pressure waves on the basis of their amplitude, duration, AUC and propagation, using various methods. Firstly, distribution plots for all mentioned characteristics lacked a bimodal pattern. In other words, none of the distributions showed an additional peak in the tail of the histogram. This finding precluded objective determination of an HAPC detection threshold for amplitude, duration or AUC. With increasing high-amplitude thresholds for HAPC detection, a gradual decrease in the automatically detected number of HAPCs was observed, again precluding determination of a specific amplitude threshold.
Furthermore, this study showed that there is a substantial degree of intraobserver variability between experienced investigators when HAPCs are identified without making use of well-defined criteria. In most previous investigations, including our own, arbitrary selected were used amplitude criteria for HAPC detection. Some investigators used amplitude criteria that were determined by taking the mean + 2 SD of the amplitudes of propagating pressure waves.9 However, this method has some important shortcomings. Firstly, this limit is dependent on the amplitude threshold for peak detection; the lower the threshold, the lower the 95% value. Secondly, as shown in this study, amplitudes of colonic contractions are not normally distributed. Instead, they have a highly asymmetrical distribution, skewed to the right. As a consequence, the mean + 2 SD does not provide a reliable estimation of the 95% confidence limit. In this study, the upper limit of normal calculated by adding 2 SD to the average was considerably lower than the values represented by Rao and Cook. The discrepancy could be explained by the fact that their analyses were performed manually.9,10,12 It is likely that the human eye misses large proportions of very low-amplitude propagated pressure waves that are detected by the computer. Another option would be to take the 95th percentile of all detected pressure waves or all detected propagated pressure waves as minimum detection value for HAPCs, but this again would be dependent on the minimum detection threshold of peak detection. Furthermore, by applying the 95th percentile, it is assumed that HAPCs belong to the upper 5% of the total range of (propagated) pressure waves detected. However, HAPCs account for a smaller proportion of the total amount of colonic pressure waves measured in a 24-h period than the upper 5%, i.e., the mean number of visually detected HAPCs (178) represented 2.0% of all propagated sequences (n = 8758). Furthermore, these 178 HAPCs consisted of about 750 pressure peaks, which represented only 0.5% of all detected peaks (n = 141.093). The 95th percentile values found in this study confirm this, because they appear to be quite low for amplitude, duration and AUC, also when only propagated pressure waves are taken into account. Any choice of a higher percentile value would again be arbitrary.
Characteristics of human colonic motility tracings are likely to be affected by the technique that is used to record them. For instance, recordings obtained with solid-state transducers have higher rise rates and higher peak amplitudes than do recordings obtained with a perfused catheter.14 This may also account for some of the discrepancies between previous studies. Furthermore, minor discrepancies in the number of events may be due to different colonic preparation. Contractile activity is greater after a cleansing enema than in the unprepared colon.15 However, we performed prolonged recordings for periods of at least 40 h, and we did not include the first day and night (excluding about 14 h), to avoid this preparation effect.
This is the first study in which automated detection of HAPCs was described and evaluated by comparing it to visual analysis. Because we have shown in this study that objective determination of a cut-off amplitude to distinguish HAPCs is not possible, visual HAPC counts by three experienced investigators were taken as reference sets. Repeated computer analysis using an array of amplitude thresholds showed that the best correspondence with the visual analysis was obtained with an amplitude threshold of 100 mmHg in two channels and 80 mmHg in one channel.
In conclusion, objective criteria that would have made it possible to distinguish HAPCs from other propagated colonic pressure waves on the basis of their amplitude, duration or AUC do not exist. However, automated detection of HAPCs using empirically derived criteria leads to an acceptable degree of correlation with visually detected HAPCs.