Presented at the Society for Academic Emergency Medicine Western Regional Research Forum, Portland, OR, March 2007; and the Society for Academic Emergency Medicine annual meeting, Chicago, IL, May 2007.
Measuring Nausea in Emergency Department Patients via the Use of Exploratory Factor Analysis
Article first published online: 2 JUN 2010
© 2010 by the Society for Academic Emergency Medicine
Academic Emergency Medicine
Volume 17, Issue 6, pages e33–e39, June 2010
How to Cite
Cloutier, R. L., Fuller, B. E., Hendrickson, R. G. and Jones, N. (2010), Measuring Nausea in Emergency Department Patients via the Use of Exploratory Factor Analysis. Academic Emergency Medicine, 17: e33–e39. doi: 10.1111/j.1553-2712.2010.00774.x
- Issue published online: 2 JUN 2010
- Article first published online: 2 JUN 2010
- Received July 15, 2009; revisions received October 16 and December 1, 2009; accepted December 6, 2009.
- emergency department;
- factor analysis
Objectives: The objective was to evaluate the applicability of a previously studied multifactorial nausea scale in the emergency department (ED) setting via exploratory factor analysis (EFA).
Methods: Two studies evaluated the validity and factor structure of 18 nausea descriptors scored on 11-point Likert scales. Trained research volunteers administered the scale to 83 men and 123 women in the first sample and to 100 men and 230 women in the second sample. All patients were assessed at enrollment and again at 90 minutes to detect changes in symptom severity. An EFA in the first study used a maximum likelihood estimation method with a principal factor analysis. The second study narrowed the descriptors and evaluated the factor structure with a confirmatory factor analysis (CFA).
Results: Two factors were retained in the solution; one contained five items with descriptors of physical symptoms, and a second contained five items with psychological symptoms. CFA determined that the two five-item scales were stable and reliable measures of patient nausea experience.
Conclusions: The scales measure both physical and psychological symptoms of nausea, indicating that the experience is multidimensional.
ACADEMIC EMERGENCY MEDICINE 2010; 17:e33–e39 © 2010 by the Society for Academic Emergency Medicine
Nausea is one of the most common complaints treated in the emergency department (ED) setting. A vast array of disease processes including motion sickness, gastrointestinal illnesses, cardiac disease, medication side effects, and intracranial disorders may all present with some degree of nausea. Successful treatment of nausea in the ED offers patients more timely symptomatic relief and less physiologic and psychological distress. Carefully tailored and documented assessment of nausea may provide practitioners with more effective means of making appropriate ongoing treatment and disposition decisions for patients experiencing nausea.
Few ED-based studies exist measuring the effects of antiemetic therapy or the experience of nausea as a symptom.1–5 In contrast, pain has multiple clinical measures that have been developed and validated within the ED setting, including visual analog scales (VAS) and numerical ranking scales (NRS) as means for quantifying the experience and severity of pain.6,7 All of these previous studies, however, have used single-item scales that are limited in their capacity to measure any given outcome.
Muth and colleagues8 developed a multi-item scale through the development of a psychometric tool measuring numerous aspects of the nausea experience as an alternative to the VAS and NRS approach. They produced the Nausea Profile, an inventory of 18 descriptors validated in an experimental model (hence in a nonclinical setting) that induced varying degrees of nausea in healthy volunteers via a rotary optokinetic device. The results were analyzed with exploratory factor analysis (EFA), which found three distinct factors: physical, psychological, and gastrointestinal. Each factor was developed in a four-stage process, which started with the use of numerous descriptors of nausea that were ultimately pared down to 18 items. The Muth Nausea Profile was validated by inducing motion sickness in half of the study volunteers. It was found that compared to those who had not been subjected to motion sickness, each of the factors was higher in those who were sickened.8 However, an 18-item scale is difficult to apply in an ED clinical setting and, to date, there are no examples of multidimensional scales in the emergency medicine (EM) literature. There are examples of factor analysis in the EM literature by Kelly et al.,9 and Houry et al.,10 Bryant et al.,11 and Barbosa et al.12 have also published methodologic discussions of confirmatory factor analysis (CFA). Thus, the major goals of this study were to 1) employ a multifactorial approach using factor analysis to evaluate the experience of nausea in ED patients as a multidimensional experience, 2) attempt to narrow the scope of the Muth scale by determining which items are of highest impact, and 3) validate the use of a shorter multifactorial scale in an ED patient population. Additionally, the reliability and validity of these items for both time periods were evaluated.
This article presents two studies validating the items from the Muth Nausea Profile to create a brief, multidimensional scale based on the experiences of ED patients presenting with nausea. Study 1 applied this set of items to a set of 206 patients between the months of August and October 2005. An EFA identified the factor structure of these items and identified two independent factors consisting of five items each. Study 2 collected an additional 330 patients with a narrower 10-item version of the scale and confirmed the validity of the factor structure between August and October 2007.
Both studies were prospective observational studies conducted via survey instruments. The Oregon Health and Science University (OHSU) Institutional Review Board approved the study prior to volunteer training and data collection. All patients provided informed consent prior to participating in the study.
Study Setting and Population
The studies were conducted in the OHSU ED, which cares for approximately 40,000 patients per year and is a widely recognized tertiary care and academic training center. Portland is a city of approximately 550,000 people (1.3 million population in the metropolitan area) located in the Pacific Northwest region of the United States.
The study included all adult patients (at least 18 years old) presenting to the OHSU ED with the following chief complaints: abdominal pain, nausea, vomiting, nausea and vomiting (combined), headache, sore throat, and fever. These symptoms were likely to include those who were nauseous as well as those who were not. In an analysis such as this it was important to have a wide variation on the experience of nausea.
Exclusion criterion included previous study enrollment, non–English-speaking, altered mental status as determined by the treating health care provider (either nurse or physician), patients in police custody, patients on psychiatric holds, and patients deemed too clinically unstable to participate by the treating health care provider (either physician or nurse). This definition was broadly defined and left entirely to clinician discretion. Patients who had received antiemetic therapy prior to obtaining the baseline study measurements were also excluded.
For both studies, trained student volunteers screened all ED patients based on their chief complaint for potential study eligibility. Patients were identified by monitoring the computerized tracking system EMSTAT (A4 Health Systems, Cary, NC; the OHSU ED used the EMSTAT electronic patient tracking system between 1997 and April 2008, and it was operationally used for recording data starting in 1998), holding discussions with the treating nurse or physician, and monitoring of the electronic medication access system (PYXSIS, CareFusion Inc., San Diego, CA) for withdrawal of antiemetic medications.
Student volunteers were members of the Clinical Research Investigative Studies Program (CRISP). CRISP students are present in the ED during the hours of 07:00 to 23:00 daily for activities related to study data collection (no data collection occurred between 23:00 and 07:00). CRISP volunteers are fully trained in all aspects of informed consent and data collection, including Health Insurance Portability and Accountability Act guidelines as they pertain to clinical research. All CRISP participants attended a 2-hour training session to become familiar with administration of the research tool and the study protocol prior to the commencement of the study.
The volunteers approached the patient during a brief interval between when he or she was identified and when that patient received first dose of antiemetic medication. Basic demographic information was obtained directly from the electronic medical record. The volunteer briefly explained the intent of the study and obtained informed consent from the patient to administer each of the scales as described below. The volunteer returned at 90 minutes to obtain a second nausea profile, and the patient was asked to rate any change in symptoms on a five-point scale (“since the last time I talked with you is your nausea vomiting:” a lot worse, a little worse, no change, a little better, a lot better). All diagnostic, therapeutic, and disposition decisions were at the sole discretion of the treating physician and were not affected by the patient’s decision to participate in the study. In routine clinical practice in the OHSU ED, patients are asked to rate their symptoms before and after treatment interventions for both nausea and pain.
Patients were ensured of the confidentially of their responses and that the various scales were not to be part of the medical record. Each survey instrument was labeled with only a study number for identification and was tracked by the CRISP volunteer. CRISP volunteers, and not the ED staff, collected all data. At the time of patient disposition, the research assistant abstracted a few limited data points from EMSTAT and recorded these data points in the study chart. Data were entered into a Microsoft Access database (Microsoft Corp., Redmond, WA) that was password-restricted with access only by the investigators. Completed data were kept in a locked file accessible to only the primary investigator or the CRISP coordinator.
The survey instrument included a set of 18 items, each scored on an 11-point Likert scale (0 to 10) describing different subjective feelings of nausea (Table 1). The items were administered at baseline, immediately following consent, and again at 90 minutes to detect changes in symptom severity. The instrument also assessed whether the patient had vomited prior to enrollment and if the patient felt different at the 90-minute interval. Two different studies were conducted using this design. The first was to validate the original 18 factors by Muth et al. in the ED setting, and the second to validate a narrower set of the highest yield factors obtained from the EFA of the first study.
|Item||Physical Discomfort||Physical Discomfort at 90 Minutes||Psychological Discomfort||Psychological Discomfort at 90 Minutes|
|I feel sick||0.84||0.83||0.14||0.24|
|I feel if I might vomit||0.80||0.65||0.01||0.02|
|I feel nauseous||0.80||0.75||0.06||0.15|
|I feel ill||0.80||0.77||0.06||0.20|
|I feel queasy||0.72||0.77||0.23||0.39|
|I feel discomfort in my stomach||0.56||0.48||0.28||0.25|
|I feel weak||0.50||0.58||0.25||0.24|
|I feel worried||−0.01||0.15||0.74||0.78|
|I feel scared or afraid||0.13||0.19||0.73||0.89|
|I feel upset||0.17||0.36||0.68||0.76|
|I feel nervous||0.09||0.11||0.66||0.76|
|I feel panicked||0.19||0.19||0.65||0.74|
|I feel shaky||0.28||0.35||0.43||0.34|
|I feel lightheaded||0.37||0.47||0.21||0.29|
|I feel fatigued or tired||0.36||0.55||0.33||0.20|
|I feel hot or warm||0.30||0.32||0.24||0.17|
|I feel sweaty||0.19||0.23||0.35||0.34|
|I feel hopeless||0.15||0.26||0.34||0.58|
Exploratory Factor Analysis. An EFA was conducted in Study 1 on the 18 items collected at the baseline measurement using the “PROC FACTOR” module in SAS (1999, SAS Institute, Cary, NC). An EFA is designed to assess a set of items measured on a Likert scale and find the common variation among the set of items. This is done by analyzing the covariance matrix of the items and extracting common variance “chunks,” referred to as eigenvalues. Eigenvalues are measured in the units of the variance of one item; thus, an eigenvalue of 2.5 would contain the variance of about two and a half items. These values are examined, and those that are over 1.0 usually represent a “factor.” With careful examination of a plot of these eigenvalues (a scree plot), the authors decide how many factors they will test for. By rotating the axes of the factors, the items will increase their relationship to one factor (increase a factor loading) and decrease their relationship on the other factor(s). This generates a simple structure such that one item loads ideally on one factor. There are two main rotation methods, orthogonal (e.g., varimax), which assumes that the two factors are not correlated, and oblique (e.g., promax), which assumes correlation. The factor is then assessed by examining the content of the items that load on it. This establishes the content and the name of the factor. This analysis used a maximum likelihood estimation method, which is a Bayesian method of calculation that is used to establish a solution that is less variable with multiple datasets. This was a principal axis factor analysis because we used the squared multiple correlation coefficient on the diagonal of the variance/covariance matrix. The EFA was the subject of Study 1. The appropriate sample size for an EFA with 18 items would be 180 (10 per each item). We had 206 in this study, which more than met this requirement. There is no penalty for being overpowered in a study such as this, as more participants lead to a more stable solution.13,14
Confirmatory Factor Analysis. A confirmatory factor analysis (CFA) was conducted to cross-validate the five highest-yield items from the physical discomfort scale, and the five highest yield items of the psychological discomfort scale. The total of 10 items was chosen a priori as a reasonable number of items for a screen used in the ED setting. A separate sample was used with the items retained from the previous analysis. This is essential for cross-validation and ensures that the data used to confirm the factor structure is different than what was used to create the original EFA. A CFA uses a structural equation model of the two factors and the items that load onto the factor. This analysis is different from the EFA because it explicitly states which item loads on which factor and does not allow for deviation or changes. If this model fits the data well, the authors can be confident that the factor structure of the original EFA is a good match for the second sample and, presumably, any future samples collected from the same population. A maximum likelihood solution was used to arrive at convergence with the program Mplus.15 The fit of the model determines how well the structural model fits the data. The comparative fit index (CFI) is a robust, unbiased measure of model fit and should have values close to 1.00. The root mean square error of approximation (RMSEA) is also used as a measure of model fit. Good models have an RMSEA of 0.05 or less; models that have an RMSEA at 0.10 or higher have poor fit.13,14 Included in this model were the two outcome variables “Did you vomit in the last 24 hours?” and “How many times did you vomit?” This allows a direct test of the relationship between the factor and these two questions. If significant, it indicates that the factor differed as a result of the question. The sample size for this analysis would also require 10 observations per item. In this model there are 10 items and two additional questions, and thus 120 subjects would adequately power this model.13,14 We had 330 subjects in our sample.
Participants. The sample consisted of 206 subjects: 83 men and 123 women. The mean age of the sample was 39 years (range = 19 to 87 years). A total of 359 patients were initially approached for participation and consent; of these, 45 patients refused participation, nine were missed, and 68 met secondary exclusion criteria (most commonly for being non–English-speaking, mental illness, or deemed too ill to participate). Thirty-one surveys were incomplete and could not be analyzed, leaving 206 subjects remaining for analysis. The racial mix of subjects was 78% white, 10% Hispanic, 3% African American, 3% Asian, and 5% undetermined.
Exploratory Factor Analysis. The initial unrotated solution found six eigenvalues above 1.0 indicating the presence of more than one unique factor. Examination of the scree plot, a plot of eigenvalues in order, showed that at most, three factors would predict the majority of the variance in these 18 items. This is done by examining the relative size differential between the eigenvalues. As the relative change decreases, the chance of a distinct factor decreases. Models with two and three factors were calculated. The items were rotated with both promax and varimax rotation methods. The best fitting model, one in which each item loaded cleanly on one axis, was achieved with a two-factor model with a varimax rotation. This rotation assumes that the two factors are not correlated. The actual correlation between the two factors was r = 0.19 confirming the independence of the two factors. We named these two factors according to the content of their items, the Physical Discomfort Scale and the Psychological Discomfort Scale.
Physical Discomfort Scale. The first factor contained the items “I feel sick”“I feel as if I may vomit,”“I feel nauseous,”“I feel ill,” and “I feel queasy” in order from the highest to lowest factor loadings. Each of these items described physical sensations. The internal consistency (coefficient alpha) of this set of items was 0.90, implying that this was a reliable scale. When these items were given 90 minutes following the first administration, the factor structure on this set of data was identical (see Table 1, second column), the coefficient alpha was 0.89, and the coefficient of stability (test–retest correlation) was 0.54. We felt this was acceptable given that nausea is a state measure (as opposed to a trait) and likely to change over time. The mean values for this score at time 1 were 18.06 and at time 2 were 13.13. Mean differences were calculated for both scales between those who answered that they had vomited in the past 24 hours prior to the survey. A repeated measures F test was used to test the hypothesis that those participants who reported vomiting (n = 89) had a significantly higher mean on the Physical Discomfort Scale (M = 23.01) than those who did not report vomiting [n = 109, M = 12.29, F(1,196) = 35.40, p < 0.0001]. Respondents were also asked at the 90-minute mark if they felt their symptoms changed during their stay at the ED. A significant relationship was observed between those who reported feeling better and the difference in their physical symptom scale from time 1 to time 2 [F = (2, 136) = 19.03, β = 0.45].
Psychological Discomfort Scale. Five items loaded onto a second factor and included “I feel worried,”“I feel scared or afraid,”“I feel upset,”“I feel nervous,” and “I feel panicked” in order of decreasing factor loadings. The coefficient alpha for this set of items was 0.845. At time 2 the factor structure was also similar and the internal consistency estimate was 0.875. The correlation between the two time points was 0.72, indicating a coefficient of stability. This shows that the psychological experience of nausea may be less malleable over time. The mean at time 1 was 11.22 out of a possible 50 and at time 2 was 7.50 out of a possible 50. The psychological discomfort scale was also significant between those who reported vomiting in the prior 24 hours (n = 94; M = 13.12) compared to those who did not [n = 108; M = 9.57, F(1,200) = 6.46, p < 0.0118]. The change in psychological symptoms did not predict change in the physical discomfort scale. The correlation between these two scales was 0.19, confirming the relative independence of the two factors. This indicated that the experience of nausea was multidimensional and involved both a physical and a psychological component.
Participants. The sample for analysis consisted of 330 patients: 100 men and 230 women. The mean age was 39 years (range = 18 to 83 years). A total of 528 patients were initially approached for participation and consent; of these, 65 patients refused participation, 13 were missed, 108 met secondary exclusion criteria (most commonly for being non–English-speaking, mental illness, or deemed too ill to participate), and 12 surveys were incomplete and could not be analyzed, leaving 330 subjects remaining for analysis.
Confirmatory Factor Analysis. The fit functions for this model are quite good and show an outstanding model fit (χ2= 111.31, df = 49; CFI = 0.974; RMSEA = 0.064). This model is presented in Figure 1 and displays standardized regression coefficients for all paths (range in value from −1.0 to 1.0). (All paths are significant at the p < 0.001 level.) One covariance between the error terms of the item “I feel sick” and “I feel ill” was included to improve model fit. This shows that these two error terms have a 0.43 correlation suggesting similarity of content. This model also shows that each factor is independent of the other, sharing only a 0.28 correlation coefficient and does not have any cross-loading items.
Physical Discomfort Scale. The Physical Discomfort Scale showed an alpha value of 0.905 for time 1 and 0.913 for time 2. The means of this scale at both time points were 23.49 and 16.22. The test–retest coefficient was 0.703. This scale predicted not only whether the patient vomited in the past 24 hours, but also the number of times he/she vomited (see Figure 1).
Psychological Discomfort Scale. The Psychological Discomfort Scale showed a high internal consistency of 0.896 for time 1 and 0.933 for time 2. The means for this scale were 16.21 and 11.90 at times 1 and 2, respectively. The test–retest coefficient was 0.774. This scale did not predict vomiting.
This is the first study in the EM literature to use factor analysis and a multi-item scale for the measurement of nausea in the ED. This study extends previous work by Muth et al.8 by applying the Nausea Profile to a clinical population and evaluating nausea as a quantifiable multidimensional experience. The study confirms a scale measuring the elements of the patient experience related to nausea as two distinct factors: the physical and the psychological. The five-item scale for physical discomfort predicts past vomiting and is sensitive to changes occurring over the time of treatment. The psychological discomfort scale measures a distinctly different experience of nausea related to fear and anxiety but does not tend to predict vomiting. These scales are independent and can be used together or separately. The factor structure does not change depending on whether one or both scales are administered. The physical scale is a direct measure of the physical symptoms of nausea; however, the psychological scale, while standing alone as a separate factor, may measure other aspects of the nausea experience. More studies are required to understand the nature and outcome of this factor.
Nausea has often seemed more elusive than pain to study in the ED due to its more abstract nature. Previous authors have applied VAS or NRS to measure nausea in general and as a means of evaluating different antiemetic treatments.2–5,16–18 The VAS and NRS are single-item scales and have significant advantages as clinical tools in the ED setting due to their simplicity and ease of use. Yet, it is the very simplicity of these single-item scales that may underlie their weakness. Many other subjective experiences such as depression present in multiple ways, and measurement of these constructs with single-item scales often produces unreliable results. In the behavioral sciences, multi-item scales have been used to measure these more subjective and abstract clinical entities. Debate continues as to the validity of single-item scales versus multi-item scales. Several recent studies have shown that single-item scales have strong correlations with multi-item measures19 and conclude that single-item scales are empirically no better than multi-item scales for measurement of single constructs.20 However, when an underlying construct is multidimensional, single-item indicators may be inadequate.
The introduction of a multidimensional scale may preliminarily now lead to the potential for studies assessing the relative merits of various treatment regimens or different antiemetic agents. We are at a point where we may be able to better assess our treatment of nausea via the development of a multidimensional tool with a scale suited for the ED environment. Clinicians potentially may use the five-item physical discomfort scale to quickly and reliably measure nausea in patients and then evaluate the effects of antiemetic treatments with a compact, sensitive instrument. The psychological scale may also allow clinicians to independently better characterize the side effect profiles of antiemetic agents (drugs such as prochlorperazine, promethazine, and metoclopramide which all possess significant akisthetic side effects) from the physical effects of nausea. Future studies may concentrate on more exacting therapy for nausea: for example, ondansetron for patients with high physical scores and low psychological scores and lorazepam for patients with high psychological components.
Future studies using factor analysis may also be applied to other more abstract symptom complexes frequently seen in the ED setting. For example, previous studies have alluded to sex differences in terms of the experience of myocardial chest pain and hence acute myocardial infarction or acute coronary syndrome. Factor analysis could potentially be applied to assess a number of subjective descriptors of cardiac pain to see if there are pertinent differences in the qualitative nature of the experience to help us sharpen our diagnostic sensitivity, particularly for women.
A small number of patients were potentially lost to enrollment as a result of volunteer failure to screen patients prior to administration of antiemetic meds. Data points were also not collected during the hours of 23:00 and 07:00 since CRISP students were not present during that time. Our results also do not control for different types of antiemetic medication that may or may not have been used for patient treatment. Therefore, it is possible our results may be driven in some manner by the use of ondansetron over prochlorperazine, for instance. Yet, the purpose of this study was not to measure the effects or potential benefit of a given antiemetic medication relative to another, but to assess the applicability of a multi-item scale in the ED setting and to establish the multidimensional nature of nausea in the ED patient. It is also important to acknowledge that the purpose of this study was not to compare our scale against previously studied scales such as the VAS or NRS. Indeed, it would be immensely valuable for future studies to assess the feasibility of using a multidimensional tool in the ED and compare it directly against a one-dimensional tool.
The potential effects of missing data we believe are minimal. In Study 1, we had 31 incomplete surveys representing 13% of the sample; in the context of factor analysis this is unlikely to bias the final outcome. This absence of effect is similar for Study 2, where 12 missing surveys were noted comprising 3.5% of the sample.
This measurement was developed and tested largely in white adults. Its applicability to minority populations and children remains unstudied. These results should be interpreted with caution and applied within the intended goal of the study, which was to demonstrate the applicability of multi-item scales in the ED for the measurement of abstract symptom constructs.
We conclude that a multi-item scale for the experience of nausea is applicable in the ED setting. The scale measures both physical and psychological symptoms of nausea as distinct factors, indicating that the experience is multidimensional.
- 2Clinically significant improvement in nausea as measured on a visual analog scale. Acad Emerg Med. 2003; 10:437–8..
- 13Structural Equations with Latent Variables. New York, NY: Wiley, 1989..
- 14Multifaceted conceptions of fit in structural equation models. In: BollenKA, LongSL, eds. Testing Structural Equation Models. Newbury Park, CA: Sage Publications, Inc., 1993..
- 15Mplus Users Guide. Los Angeles, CA: Muthén & Muthén, 1998., .