Statistical Analysis of Total Labor Pain Using the Visual Analog Scale and Application to Studies of Analgesic Effectiveness During Childbirth

Abstract

There is a need to assess total pain in labor studies. Typically, studies measure pain in 30-min intervals or at specific degrees of cervical dilation [1-3]. Resulting pain measurements reflect the amount of pain the woman feels at these specific times. Because labor pain constantly changes, these measurements may not accurately reflect the overall intensity or duration of pain throughout labor [4]. For example, even if one drug were to reduce pain more than another drug for some period of time, if it were also to substantially prolong the total duration of labor, then women may experience greater total pain. Alternatively, the analgesic effect of a drug may begin to lessen over time, providing less analgesia for longer than for shorter labors. Because total pain measurements account for differences in the overall duration of labor among women, they give a more accurate representation of analgesic relief than do individual pain measurements. The purpose of this article is to review statistical issues pertaining to the measurement of total labor pain. Because we are not aware of data regarding patients' relative satisfaction with treatment of intense pain over a brief duration versus mild pain over a protracted duration, we do not include consideration of patient satisfaction. The assessment of the total pain experienced throughout labor is vital in cost-effectiveness analyses, in which the ratio of costs to benefits of a new treatment are compared with that of an existing treatment [5]. For example, the relative cost of administering ropivacaine via an epidural catheter versus sufentanil via a single-shot spinal would be compared with the relative benefits of each drug. The measurable benefits of each drug, including magnitude and duration of pain relief, would be discounted by their cost. If one drug relieves the same amount of pain over a longer period but the total cost of administering it is less, it would be more efficient according to cost-effectiveness analysis [5]. If a drug provides more pain relief but prolongs the duration of labor, the costs of administering the drug would include the direct and indirect costs associated with the prolongation of the duration of labor [5]. Determination of the cost-effectiveness of a labor analgesic requires that the magnitude of pain relief and duration of pain relief be combined (i.e., that total pain relief throughout labor be quantified). When a drug is considered for use in a certain clinical arena, it is tested in clinical trials to determine its toxicity and efficacy. For example, in studies of labor analgesics, a visual analog scale (VAS) score may be obtained, a drug then administered via a single-shot spinal, and another VAS score then obtained. The difference between the VAS scores would provide information as to whether the drug administered spinally has efficacy as an analgesic. This means that the clinical trial would determine whether the drug can provide analgesia. After efficacy has been established, effectiveness studies would be performed to determine whether the drug administered via a single-shot spinal is an effective analgesic relative to other available analgesic modalities when the drug is used in routine clinical practice. How the drug is used in clinical practice can therefore have a large effect on the drug's cost-effectiveness. For example, if a drug that provides a long duration of analgesia after being administered via a single-shot spinal were developed, the cost-effectiveness of the drug would be greater if it were routinely administered early in labor rather than just before delivery. The issue that we address in this article is how to quantify analgesic effectiveness during childbirth. Pain Measurement Theory Researchers have yet to identify the best way to measure the magnitude of acute pain. Asking a series of questions regarding different aspects of pain has some advantages because pain has both sensory and affective components [6]. Other researchers use simple verbal descriptions such as "nil," "some," or "excruciating" [7]. Physician or nurse observers have also been used to estimate a woman's intensity of pain [2]. The VAS is reliable and valid in the measurement of pain [8-11], can detect smaller differences in pain than other scales (such as a five-item verbal descriptor), and is easier to administer than a series of questions [12-14]. The VAS is a vertical or horizontal line, usually 10 cm in length, with descriptive words such as "no pain" and "worst pain imaginable" written at the end points [11]. The patient is asked to mark the line at the point that they consider represents the amount of pain they are experiencing at that moment [11]. The distance from the end point to the mark, often measured to the nearest 0.5 cm, is recorded as their pain score [15]. Methods of pain relief are then often compared by analysis of the resulting pain scores [15]. One of the most important aspects of the validity of the VAS is that it is a quantitative measure of pain. A VAS score of 0 cm represents a true zero because it is an indication of the absence of pain. This means that VAS measurements are ratios in nature, meaning that a score of 6 cm is twice as severe as a score of 3 cm. The difference between VAS scores of 2 and 3 is of the same magnitude as the difference between VAS scores of 7 and 8. Being a ratio level of measurement, the VAS score can be multiplied by other factors, such as the duration of labor, which allows a total pain score to be calculated for each subject. Total pain relief can be obtained using the VAS but not with verbal descriptors of pain. Inclusion Criteria for Studies of Labor Pain Throughout the course of any study, problems in data collection may lead to missing values. Statisticians characteristically divide missing values into two types: informative and noninformative. Missing data are informative if they are directly related to the value that would have been observed, such as if an observation is missing because the woman was in so much pain that she refused to complete the VAS. Missing data are noninformative if the data are missing for reasons that are not related to the woman's level of pain, such as if the data collector forgot to ask the woman to complete the VAS after the third hour of labor. The distinction between informative and noninformative missing data must be made in studies of labor pain because informative missing values are likely to cause difficulties in the statistical analysis of the pain scores. Missing pain scores are problematic in labor studies using the VAS. One of the concerns with using the VAS at designated labor time points is that women arrive at the hospital during different phases of labor [1]. For example, among primiparous women, 74% arrive at the hospital at a cervical dilation at 0-3 cm, but only 16% arrive at 4-6 cm dilation [1]. Among multiparous women, 19% arrive at the hospital at 4-6 cm and 20% arrive at 7-10 cm dilation [1]. If a researcher plans to measure pain starting when the parturient is dilated 3 cm, informative missing values will result from women who arrive at the hospital after their labor has progressed beyond this degree of dilation. These women cannot simply be excluded from a study because when a woman arrives at the hospital is related to her magnitude and duration of pain. Missing pain scores are particularly problematic for cost-effectiveness studies. Total labor pain is based on a woman's total pain throughout childbirth. Excluding from an effectiveness study those women who present to the hospital later in labor (i.e., with missing pain scores) would make cost-effectiveness results very difficult to interpret. Worst Pain Imaginable Another problem encountered in labor pain studies is the interpretation of worst pain imaginable, the descriptive endpoint of the VAS. A patient's perception of worst pain imaginable may change over the course of childbirth. For example, 13% of primiparous women without analgesia report the maximal worst pain imaginable at 3 cm cervical dilation [16]. Of primiparous patients without analgesia, 31% score the intensity of their pain between 9 and 10 cm during the first phase of labor, with 10 cm being the value corresponding to worst pain imaginable. Some women have reported pain scores of 10 cm on the VAS scale at <or=to3 cm of cervical dilation [1]. If these patients receive no analgesia, their pain will increase. Because they have already reached their maximal pain score, they will not be able to report that increase using the VAS. A VAS score of 10 cm marked by a woman with a cervical dilation of 0-3 cm may represent a different intensity of pain than an identical VAS score of 10 cm at a cervical dilation of 7-10 cm. There is no easy way to control for these changes in perception when measuring pain using the VAS over time. Overcoming Missing Values Missing values can result during a labor pain study because of factors that the investigator may have difficulty in controlling, such as if a patient is asleep. This is not an uncommon occurrence during early labor or after the placement of a well functioning epidural catheter. The investigator could consider VAS scores to equal 0 cm when women in labor are asleep [17]. However, we are not aware of a study that has validated this approach. Patients can be awakened, but this strategy raises concerns regarding patients' comfort [18]. Another tactic to compensate for such missing values is to use the mean of other women's VAS scores as an estimate for a missing value; however, this may not be accurate. The use of other patients' scores as an estimate may alter results because pain scores may indeed be lower for sleeping patients than for women who are awake. Observations may be missing in labor studies if a woman is in so much pain that she does not want to complete a VAS [6]. In postoperative pain studies, missing values can often be assumed to be missing at random and noninformative. For example values may be missing for patients who are undergoing physical therapy, walking, or receiving a bath. However, methods to predict missing at random responses are not likely to be appropriate in labor pain studies [19]. If a woman has poor analgesia during phase II, is screaming with each contraction, and refuses to fill out the VAS, using the mean of other women's scores as an estimate may be inaccurate. This method for estimating her pain also assumes that a mean value can be calculated. Because the duration of labor varies among women, VAS measurements taken at the same time since randomization into a study or since the start of labor would correspond to different cervical dilations. VAS measurements should therefore not be compared in this manner. Scores could be aligned based on cervical dilation, but this strategy would require that cervical dilation be measured at constant intervals (e.g., every 30 min). Statistical analysis of labor pain scores cannot easily compensate for missing observations. One approach that may help to overcome the difficulties of missing observations is to have an observer predict a woman's responses. This could be achieved by having one of the patient's physicians or nurses estimate the patient's intensity of treated or untreated pain [12,15,20,21]. Measurements could then be taken at specified time points throughout labor. The likelihood of missing values due to noncompliance of the patient would be decreased by using this method, making the results more valid. The problem inherent with this approach is that there will be error in the observer's measurements. Observers typically underestimate pain scores by as much as 4.5 cm [15]. It is not known whether changes in patients' expressions of the worst pain imaginable over the course of childbirth play a role in generating this large error. To make matters worse, the error is not constant over the range of possible VAS scores [12,15,20]. Patients' scores of 0-2 cm have corresponding observer scores of 0-4 cm, and patients' scores of 9-10 cm have corresponding observer scores of 5-10 cm [12,15]. An observer's score of 6 cm could either be correct, an overestimate of an actual score of 4 cm, or an underestimate of an actual score of 8 cm [15]. If an observer's score is considered correct when it is within 1 cm of the patient's score, >65% of observers' scores are reported to be incorrect [15]. Patients consider a 1-cm change in VAS score to be clinically important [22]. Therefore, the magnitude of the difference between the observers' and patients' VAS scores could be sufficient to contribute to clinically important differences in a study's results and conclusions. Having patients record their average labor pain soon after delivery is another way to minimize the number of missing values [8]. Using this approach, only one pain measurement has to be taken for each woman, which reduces the chance of missing values. A women would not have to be awakened or asked to complete the VAS at a difficult moment. One potential criticism of this approach of women estimating their own average pain is bias related to recalled observations. There may be nonconstant bias in a patient's recalled measurements of acute pain, just as there is in an observer's measurements. Although this potential bias should be considered, it has not actually been detected [8,13,23]. In one labor study, on the day after delivery, women recalled their pain before and after the injection of epidural morphine or bupivacaine [8]. Just as with observers, over- and underestimation may have occurred but, if so, with less frequency and smaller magnitude of error than made by observers [8]. In fact, 90% of the patients' scores were within 1 cm of the actual score, and >70% were correct within 0.5 cm [8]. Initial scores of 0-1.7 cm had corresponding recalled scores of 0-1.2 cm. Initial scores of 7.1-9.9 cm had corresponding recalled scores of 6.9-8.7 cm. Furthermore, a recalled score of 6 cm corresponded to actual scores of 5.8-6.5 cm, a much smaller range than that for observer recorded pain scores [8,15]. The optimal time interval between labor and the retrospective measurement must be determined. Pain recall after 1 day is accurate [8,11,13,23], but an interval of 1 yr is too long to maintain accuracy [24]. Even an interval of 48 h may yield inaccurate results [25]. Although more investigation should be performed with labor pain before suggesting a rigid interval standard, a reasonable recommendation is that investigators have all women judge their average labor pain at a set time point after delivery (e.g., 2 h). A labor pain study using retrospective measurements has been interpreted to contradict these findings [25]. In this study, patients were initially asked to measure the amount of "overall pain or discomfort" they were feeling. Then, in the retrospective measurement, they were asked separately about "overall pain" and "overall discomfort" [25]. However, these retrospective pain measurements were completed <or=to48 h after the event, a longer interval than the 1 day [11,13,23] or less [8] of other retrospective pain studies. This study [25] should therefore not be used to refute evidence in support of retrospective measurements. Rather, it reinforces the concept that the interval between delivery and pain scoring should be minimized. Given the evidence, the use of retrospective VAS scores of average pain seems to be the best approach to measure total labor pain. Statistical Analysis The VAS scores from retrospective measurements represent the average pain that women experience during labor. Total pain scores can then be obtained by multiplying the VAS score by the duration of labor (phase I and phase II of labor combined). This adjustment should be applied to each pain score before comparing the total labor pain among groups. VAS pain scores themselves can be analyzed by using parametric analyses provided that there are not too many values clustered near the extreme values [16]. For example, if <13% of the study population reports a score of 0 or 10 cm, then parametric analyses such as Student's t-tests and analysis of variance are acceptable [16]. If the data are too highly clustered toward the extremes, such as >or=to25% of patients with a score of 10 cm, then nonparametric analyses should be considered. Nevertheless, when the retrospective VAS score is used to estimate average labor pain, parametric analyses would be expected to be acceptable. Because the retrospective VAS measures average pain from the beginning to the end of labor, it would not be expected that >13% of the subjects would report the maximal or minimal score [1,23]. Sample Size Necessary sample sizes for studies dealing with average pain can be estimated from the literature. The standard deviations of VAS pain scores are needed for the calculations. Because standard deviations of 1.0 cm are common for labor pain measurements [8], this value can be used for sample size calculations. Additionally, the minimal detectable difference must be determined. Patients consider a 1.0-cm change in VAS score to be clinically important [26]. Using the standard deviations of VAS pain scores measured in labor and based on the usual parametric methods (Student's t-tests and analysis of variance), 17 or 21 patients in each of two or three groups, respectively, would be needed to detect the differences in average pain that patients consider to be clinically important. Covariates The level of labor pain a woman feels may be related to factors other than her actual pain of delivery. These variables, also known as covariates, should be taken into consideration and adjusted for in the study of labor pain. First, because oxytocin may produce more pain [16], the population of patients receiving this drug should be noted. Women receiving oxytocin may need to be analyzed separately [16], or the dose of oxytocin may need to be included as a covariate [27]. Second, patients may report retrospectively larger average mean pain levels if their current level of pain is high, whereas patients with a low current pain level may underestimate their mean pain [25,28]. However, this issue should be studied further before any conclusions can be drawn with respect to labor pain. Third, educational status may also affect the estimates of pain; however, care should be taken to control for socioeconomic status when assessing the effect of education [2]. Cultural background (i.e., race) has not been shown to be an important covariate in the study of labor pain [2]. Pain decreases between phase I and phase II for primiparous women and increases for multiparous women [6]. This result suggests that the interaction of phase and parity may be important. However, because the retrospective VAS measures average pain, not pain across phases, parity may not need to be included in labor pain analyses using retrospective pain measurements. Women of different parity can probably be combined in a study with an end point of average pain [6,25]. Although covariates such as cultural background and parity may not need to be considered in the measurement of labor pain, they may need to be considered in cost-effectiveness analyses. How a labor analgesic is used in clinical practice can affect a drug's cost-effectiveness. Cultural background or parity may affect the time from when a women arrives at the hospital and receives a labor analgesic to when she delivers. For example, two drugs that provide the same magnitude of analgesia are to be compared, but the more expensive drug provides a longer duration of analgesia than the less expensive drug. The relative cost-effectiveness of the two drugs depends on the expected duration of labor at the time at which the drugs are administered. Cost-effectiveness of a labor analgesic drug or treatment may need to be stratified based on covariates such as cultural background or parity. If an investigator were to decide to stratify their analysis, then the sample sizes that we previously discussed would need to be increased accordingly. In summary, labor pain is an important aspect of medical science that needs to be further investigated. The increasing focus of healthcare research on cost-effectiveness analyses requires that total labor pain, in addition to average labor pain, be considered. Complicating this research is the problem of how to accurately measure labor pain. A compromise between accuracy and the ability to effectively analyze results must be made. The single retrospective VAS score, although not perfect, may give accurate results that can be analyzed easily. Studies may be able to include total labor pain by multiplying the retrospective VAS score by the duration of labor.