Occurring in approximately one-third of untreated surgical patients,1,2 postoperative nausea and vomiting (PONV) may result in surgical wound dehiscence, delayed hospital stay, unanticipated hospital admissions, and increased aspiration rates. PONV prevention is therefore important to improve patient safety and medical outcomes, enhance patient satisfaction, and contain medical costs.
Risk factors predicting the incidence of PONV include sex, history of PONV, gynecologic (GYN) procedure, nonsmoker, and age younger than 40 years; nonsmoking female patients undergoing GYN procedures have up to 75% risk of experiencing PONV.3–5 Multiple trials leading to an array of pharmacologic therapies targeting the multifactorial causes of PONV6,7 have shown a reduction in its incidence by approximately 20% per therapeutic modality.4
Two other proposed therapies include fluid administration and carbohydrate loading. Although results are conflicting, several studies have investigated the use of perioperative fluid therapy and its effect on the incidence of PONV.8–17 A study by Dagher et al.8 showed that administration of crystalloid did not decrease PONV and antiemetic use, whereas that of Haentjens et al.9 demonstrated that intraoperative hydration with colloid solution had a minimal effect on the incidence of PONV when compared with normal saline solution. In a review of perioperative fluid management, Chappell et al.10 concluded, however, that although current evidence lacks strength, IV fluid therapy in liberal doses may reduce PONV in outpatients undergoing minor surgery.10–14
Oral D glucose has been used through the ages for the symptomatic relief of nausea and vomiting but clinical studies show mixed results.18–20 The exact mechanism has not been determined, but has been thought to be due to a direct local action on the wall of the gastrointestinal tract that reduces muscle contraction because of the high osmotic pressure exerted by the simple sugar.19 Lauwick et al.20 examined the effects of oral dextrose loading 2 hours before induction of general anesthesia and reported it to have no effect on the rate of PONV. Cook et al.21 and McCaul et al.14 investigated the use of preoperative IV hydration combined with caloric supplementation; the former authors failed to demonstrate a reduction in PONV incidence in the recovery room, whereas the latter showed an increased incidence.
In our surgical suite, there is a practice of administering IV boluses of dextrose 5% in Ringer lactate (D5LR) solution in the postanesthesia care unit (PACU) to decrease PONV. We therefore designed a controlled trial to evaluate whether the postoperative IV administration of dextrose decreases PONV in a patient population that is at highest risk: female, nonsmoking patients undergoing GYN procedures.
ASA physical status I or II outpatients, aged 18 to 70 years, scheduled for GYN laparoscopic or hysteroscopic surgery requiring general anesthesia were invited to participate in the study. IRB approval and informed written consent from each participant were obtained. Patients with a history of diabetes mellitus, congestive heart failure, renal insufficiency or failure, or currently receiving steroids or antiemetics, and those who were pregnant, were excluded. In the preoperative holding area, patients’ demographic data and state anxiety scores were collected. The State Anxiety Index is a widely used and well validated self-report anxiety assessment instrument.22–24 To confirm baseline normoglycemia, preoperative blood glucose levels were obtained during IV placement using a portable glucometer (SureStepFlexx; LifeScan, Johnson & Johnson Co., Milpitas, CA).
Participants were randomized into 2 groups: the treatment group received D5LR solution, and the control (placebo) group received Ringer lactate (LR) solution. Randomization was performed using the permuted block method with pharmacy-controlled allocation concealment. In the PACU, the intraoperative fluid solution was discontinued and the randomized fluid was initiated. All patients received 1 L of this fluid over 30 minutes. One dose of antiemetic medication (IV ondansetron 4 mg) was administered to all patients a half hour before emergence. Intervention solution bags were placed in sequentially numbered, sealed, opaque black plastic bags to conceal group assignment from the subject, the anesthesia provider, and the PACU caregiver.
All subjects underwent a standardized oxygen-sevoflurane-vecuronium and opioid general anesthesia. Anesthesia was induced with propofol (1.5–2 mg/kg), fentanyl (1–2 µg/kg) and vecuronium (0.05–0.1 mg/kg) to facilitate endotracheal intubation. Anesthesia was maintained with oxygen/air (1:2 L/min) and sevoflurane (1–1.5 minimum alveolar concentration). IV fentanyl and morphine were administered, in 25-µg and 1- to 2-mg dose increments, respectively, depending on the change in the patient’s heart rate and arterial blood pressure from baseline. All patients received LR solution during pre- and intraoperative periods.
On arrival to the PACU, patients’ nausea levels were assessed using an 11-point verbal rating scale (VRS), from 0 to 10, with 0 representing “no nausea” and 10 representing “the worst nausea ever.” Nausea VRS scores were obtained every half hour until discharge from the PACU. Standardized rescue antiemetic medications were administered depending on their VRS scores. Patients received no rescue medication when the VRS score was 0 to 2; 1 dose of rescue medication when the VRS score was 3 to 6; and ≥2 doses for nausea score of 7 to 10, to reduce the nausea score to 0 to 2. First-line rescue medications included (in the following order of administration) IV dimenhydrinate 25 mg, ondansetron 4 mg, dimenhydrinate 25 mg, and promethazine 12.5 mg. If initial rescue medications failed, patients were then given IV droperidol 0.625 to 1.25 mg, dexamethasone 4 to 8 mg, or were admitted overnight if refractory to treatment.
A standardized PACU protocol for pain control was used. Patients’ pain was rated by a verbal numeric pain rating scale, an 11-point scale from 0 to 10, whereby 0 is “no pain,” and 10 the “most intense pain imaginable.” Patients were given no fentanyl when the pain (numeric pain rating scale) score was 0 to 3, 25 µg of fentanyl for a pain score of 4 to 6, and 50 µg of fentanyl for a pain score of 7 to 10. If patients required >100 µg of fentanyl, analgesia was then supplemented with morphine or hydromorphone in increments of 1 to 2 mg or 0.1 to 0.2 mg, respectively, until the pain score decreased to 0 to 3. The end point of treatment was to reduce the pain score to 0 to 3. The total number of doses of antiemetics used, amount of analgesics consumed, and episodes of nausea and/or vomiting were obtained every half hour from time of arrival to the PACU until discharge. To avoid interrater variability, an independent observer-investigator who was blinded to group assignment collected data in the PACU.
The primary objective of this study was to compare the incidence of PONV (Yes/No) in the study treatment groups. An a priori power analysis suggested that a sample size of 32 in each group (2 groups) yielding a total size of 64 achieved a power of 0.8 (α = 0.05) to detect a difference of 33.2% in the incidence of PONV assuming a baseline incidence of 65% in patients under similar GYN laparoscopic/hysteroscopic procedures.25,26
The secondary efficacy measures included measurement of nausea VRS scores, antiemetic medication consumption, and PACU length of stay (LOS). The antiemetic medication consumption was defined as all antiemetic medication doses consumed by each patient for the entire PACU stay. Equivalency of LOS with management of PONV was defined as a discharge time difference of ≤10% between the 2 groups. Opioid analgesics consumed (fentanyl, morphine, or hydromorphone) were all converted to IV morphine milligram-equivalents.
Continuous measurements were assessed for normality of distribution by the Shapiro-Wilk test with the chosen α level of P value 0.05. When not normally distributed, data were summarized as a median and interquartile range; otherwise, a mean and standard deviation were given. Normal continuous data were compared by using 2-sample Welch t test (or unequal variance t test) and non-normal continuous data were compared by using Wilcoxon rank sum test. All categorical data were reported as number and percentage and were compared using the Fisher exact test. Percentage difference in categorical data between 2 groups and its 95% confidence interval (CI) were reported as effect size. For all continuous data, as described in recent studies,27–30 we first calculated the logarithm of the ratio R of mean difference between treatment and control groups and its 95% CI, whose values were transformed by exponential function and reported as effect sizes. The logarithm only transforms the distribution of a ratio toward symmetry and no assumption is made of a lognormal distribution. This ratio mean estimate is unit-free and can be pooled together in meta-analysis.29 The relationships between PONV VRS score at different time points and rescue antiemetics or patients’ LOS were assessed using Spearman rank correlation coefficients.
All statistical analyses were performed with the use of SAS 9.2 (SAS Institute, Cary, NC) with original database prepared by the IBM SPSS statistical software (Chicago, IL). All statistical tests were 2-tailed. The Bonferroni procedure was used to adjust P values only where between-group comparisons have been made repeatedly over time. P values of <0.05 were considered to indicate statistical significance, unless otherwise specified. Aside from the cutoff P value, a 95% CI that does not embrace the value of no difference between treatment and control groups (e.g., 1 for ratio R of mean difference) also indicates that the treatment is significantly different from the control at the 5% level. In addition, the CI provides very useful information for the range of observed effect size that helps in assessing clinical significance of research findings.31
Sixty-four patients were enrolled in the study. Two patients were withdrawn after randomization because of extreme anxiety and abnormal blood glucose. Baseline demographic and personality characteristics of the participants are presented in Table 1. According to the results of the Shapiro-Wilk normality test, the variables IV fluids total, anesthetic duration, and estimated blood loss were not normally distributed (P < 0.05), and the variables age, weight, height, state anxiety status, preoperative blood glucose, intraoperative opioid, and IV fluids in the operating room were normally distributed (P > 0.05). The last column of Table 1 lists the P values for categorical variables (Fisher exact test), non-normal variables (Wilcoxon rank sum test), and normal variable (Welch t test). Both groups were similar with regard to age, weight and height, state anxiety, and whether they had previous surgery, or had a history of PONV/motion sickness. Preoperative glucose levels, intraoperative narcotic consumption, total time in surgery, and IV fluids received did not differ between the groups (P value from 0.24 to 1). Both groups were also homogeneous with regard to surgical procedure types P = 0.62).
Table 2 presents the results of between-group comparisons of primary and secondary outcomes. No statistical significance was found for a difference of our primary end point, incidence of PONV (Yes/No) between the 2 groups (Fisher exact test, P = 0.22). Fourteen of 30 participants (46.7%) in the D5LR group experienced nausea, versus 20 of 32 participants (62.5%) in the LR group (percentage difference, −15.8%; 95% CI, −43.6% to 11.8%; Fisher exact test, P = 0.31), whereas 4 of 30 participants (13.3%) in the D5LR group experienced vomiting, versus 4 of 32 participants (12.5%) in the LR group (percentage difference, 0.8%; 95% CI, −16.7% to 18.4%; Fisher exact test, P = 1.00). However, of those participants who vomited, there were fewer total episodes in the D5LR group (5 episodes among 4 patients) than those (11 episodes among 4 patients) in the LR group.
Twelve of 30 participants (40.0%) in the D5LR group versus 18 of 32 participants (56.3%) in the LR group needed rescue antiemetic (percentage difference,16.3%; 95% CI, −44.0% to 11.5%; Fisher exact test, P = 0.22). Notably, of the patients who required rescue treatment in the PACU, the antiemetic rescue medication dose consumed by patients in the D5LR group was less than that in the LR group (ratio mean difference, 0.56; 95% CI, 0.39–0.82; Wilcoxon rank sum test, P = 0.02), i.e., dextrose in the treatment group reduced the rescue medication dose by 44%. LOS was determined, and patients in the D5LR group had a modestly shorter stay in the PACU (148 ± 52 minutes) compared with that (184 ± 74 minutes) in patients in the LR group (ratio mean difference, 0.80; 95% CI, 0.66–0.97; Welch t test, P = 0.03), i.e., a 20% reduction in LOS.
Postoperatively, PONV VRS scores were recorded beginning on arrival in the PACU and every 30 minutes thereafter. No statistical significance between the 2 groups was found after Bonferroni correction for repeated measurements of VRS over time (Wilcoxon rank sum test, P ≥ 0.18, corrected for tests at 6 time points, Table 2). There were no unplanned admissions because of PONV.
We identified a positive correlation between rescue antiemetic dose and PONV VRS scores at different time points (Spearman rank correlation coefficient, ρ > 0.40, P < 0.001), between LOS and VRS score from at 60 minutes to at discharge time (ρ > 0.40, P < 0.001). Specifically at the time of discharge, the correlations between VRS scores and rescue antiemetics, and between VRS scores and LOS, were 0.67 and 0.56, respectively (P < 0.001). A scatter plot of rescue antiemetics and LOS, 2 variables shown to have reduced measurements in the D5LR group, is shown in Figure 1.
PONV has many significant effects on both medical outcomes and the final cost for a given procedure. Furthermore, the most common prophylactic modalities are based on treating patients with additional medications, exposing them to more potential side effects. In contrast, treatment of nausea with oral dextrose (Emetrol) has been a well-established remedy associated with very low risk for nondiabetic patients.19
Pharmacologic PONV treatment comes with financial cost. Our study demonstrates a novel effect of perioperative administration of dextrose-containing fluid in reducing the number of antiemetic doses (44% reduction) required in the PACU in healthy female patients undergoing GYN hysteroscopic and laparoscopic procedures. The sole difference between the LR and D5LR group was the addition of 50 g of dextrose in the fluid bolus.
Additionally, time spent by the patient or LOS in the PACU is both onerous to the patient and costly to the care-providing institution. The current study demonstrates a 20% reduction in LOS in patients from the D5LR group compared with those in the LR group. However, whether this reduction would be seen in a nonacademic center is also worth investigating.
From our data, we identified positive correlations between PONV VRS scores and rescue antiemetic, or LOS. Although this suggests the high convergent validity of VRS as a tool of rating nausea to other efficacy measures (e.g., rescue antiemetic use, LOS) of dextrose, the choice of a best measure as a primary end point for controlled trials similar to ours is still debatable. For example, we only found the number of antiemetics (a secondary outcome) required to achieve the same VRS score was reduced in the D5LR group. This secondary outcome, despite its importance, is nevertheless not the absolute proof of the dextrose intervention.
Nonetheless, there are a few limitations in our study. First, we purposefully selected patients at highest risk for PONV and enrolled only healthy, nondiabetic, nonsmoking female patients undergoing relatively minor hysteroscopic and laparoscopic procedures. Although this produced, albeit with a relatively small study population, certain meaningful reductions in the time to discharge and rescue dosing with our intervention, it, however, was also perhaps a major limitation, that of generalizability. Extrapolating the results of this study to patients with major medical comorbidities, or those who are undergoing major abdominal or thoracic procedures, would be tenuous. Certainly, the implementation of this therapeutic modality in diabetic patients would be questionable. That being said, these data suggest that further study of the effects of dextrose-containing fluid in other nondiabetic populations that are less susceptible to PONV is alluring.
A second limitation of this study is its sample size. Our current study produced results that were interesting in terms of PONV incidence reduction but were not significant most likely because of the relatively small number of study subjects. A third limitation of this study is the measurement of a subject’s blood glucose only before surgery and not after the completion of IV fluid therapy. Whether the positive effects of dextrose administration on PONV were correlated solely to simple caloric supplementation, or to an otherwise more complex mechanism that such treatment provides, would be beyond the scope of this study and requires further investigation.
Finally, we applied Bonferroni correction for multiple testings only for VRS score, which was measured at 6 time points. Our findings on reduced antiemetic rescue medication requirement and LOS in the D5LR group would appear statistically nonsignificant (P > 0.05) if we extend the correction to all secondary end points. However, this statistical nonsignificance (by P value) does not necessarily mean clinically as “no-effect. ” CIs of observed differences are much more useful than simple P values in interpreting research findings.30,31 In our study, a ratio mean difference of 0.56 with a 95% CI ranging from 0.39 to 0.82 was observed for antiemetic rescue. Thus, the true reduction may be as much as 61% or as little as 28%, and we are 95% certain that overall antiemetic rescue dose is reduced in the D5LR group. The relatively wide range of this 95% CI is probably attributable to the small sample size of the current study. Similarly, we are 95% certain that the overall LOS was reduced in the D5LR group with the true reduction possibly being as much as 34% or as little as 3%.
In summary, under the conditions of this study, perioperative administration of dextrose-containing IV fluids reduced the number of antiemetics required and shortened the LOS in the PACU in healthy female patients undergoing GYN hysteroscopic and laparoscopic procedures. This form of PONV therapy has a low side effect profile, is easily accessible, and is inexpensive. However, further larger studies are required to confirm our results, expand to other patient populations, and to elucidate a mechanism by which dextrose may exert this antiemetic effect. E
Name: Susan Dabu-Bondoc, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Susan Dabu-Bondoc has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.
Name: Nalini Vadivelu, MD.
Contribution: This author helped write the manuscript and collect the data.
Attestation: Nalini Vadivelu has seen the original study data and approved the final manuscript.
Name: Chantelle Shimono, MSN, APRN.
Contribution: This author helped design the study and write the IRB (HIC) protocol, write the manuscript, and collect the data.
Attestation: Chantelle Shimono has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Annette English, PharmD.
Contribution: This author helped write the manuscript and perform randomization/group allocation-concealment and preparation of intervention solutions.
Attestation: Annette English has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Boonsri Kosarussavadi, MD.
Contribution: This author helped with data collection and editing.
Attestation: Boonsri Kosarussavadi has seen the original study data and approved the final manuscript.
Name: Feng Dai, PhD.
Contribution: This author helped analyze the data and edit the manuscript.
Attestation: Feng Dai has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Kirk Shelley, MD, PhD.
Contribution: This author helped design the study and edit the manuscript.
Attestation: Kirk Shelley has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Jessica Feinleib, MD, PhD.
Contribution: This author helped design the study and prepare IRB (HIC) protocol, analyze the data, and write the manuscript.
Attestation: Jessica Feinleib has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
This manuscript was handled by: Peter S. A. Glass, MB, ChB.
We are grateful to all the study participants, nurses, and other staff who made the study possible. We thank our editor and reviewers for their insights and comments that greatly improved the manuscript.
1. Gan TJ. Postoperative nausea and vomiting: can it be eliminated? JAMA. 2002;287:1233–6
2. Watcha MF, White PF. Postoperative nausea and vomiting: its etiology, treatment, and prevention. Anesthesiology. 1992;77:162–84
3. Apfel CC, Korttila K, Abdalla M, Kerger H, Turan A, Vedder I, Zernak C, Danner K, Jokela R, Pocock SJ, Trenkler S, Kredel M, Biedler A, Sessler DI, Roewer NIMPACT Investigators.. A factorial trial of six interventions for the prevention of postoperative nausea and vomiting. N Engl J Med. 2004;350:2441–51
4. Haigh CG, Kaplan LA, Durham JM, Dupeyron JP, Harmer M, Kenny GN. Nausea and vomiting after gynaecological surgery: a meta-analysis of factors affecting their incidence. Br J Anaesth. 1993;71:517–22
5. McKenzie R, Kovac A, O’Connor T, Duncalf D, Angel J, Gratz I, Tolpin E, McLeskey C, Joslyn A. Comparison of ondansetron versus placebo to prevent postoperative nausea and vomiting in women undergoing ambulatory gynecologic surgery. Anesthesiology. 1993;78:21–8
6. Kovac AL. Prevention and treatment of postoperative nausea and vomiting. Drugs. 2000;59:213–43
7. Gan TJ. Risk factors for postoperative nausea and vomiting. Anesth Analg. 2006;102:1884–98
8. Dagher CF, Abboud B, Richa F, Abouzeid H, El-Khoury C, Doumit C, Yaghi C, Yazbeck P. Effect of intravenous crystalloid infusion on postoperative nausea and vomiting after thyroidectomy: a prospective, randomized, controlled study. Eur J Anaesthesiol. 2009;26:188–91
9. Haentjens LL, Ghoundiwal D, Touhiri K, Renard M, Engelman E, Anaf V, Simon P, Barvais L, Ickx BE. Does infusion of colloid influence the occurrence of postoperative nausea and vomiting after elective surgery in women? Anesth Analg. 2009;108:1788–93
10. Chappell D, Jacob M, Hofmann-Kiefer K, Conzen P, Rehm M. A rational approach to perioperative fluid management. Anesthesiology. 2008;109:723–40
11. Maharaj CH, Kallam SR, Malik A, Hassett P, Grady D, Laffey JG. Preoperative intravenous fluid therapy decreases postoperative nausea and pain in high risk patients. Anesth Analg. 2005;100:675–82
12. Magner JJ, McCaul C, Carton E, Gardiner J, Buggy D. Effect of intraoperative intravenous crystalloid infusion on postoperative nausea and vomiting after gynaecological laparoscopy: comparison of 30 and 10 ml kg(-1). Br J Anaesth. 2004;93:381–5
13. Holte K, Kristensen BB, Valentiner L, Foss NB, Husted H, Kehlet H. Liberal versus restrictive fluid management in knee arthroplasty: a randomized, double-blind study. Anesth Analg. 2007;105:465–74
14. McCaul C, Moran C, O’Cronin D, Naughton F, Geary M, Carton E, Gardiner J. Intravenous fluid loading with or without supplementary dextrose does not prevent nausea, vomiting and pain after laparoscopy. Can J Anaesth. 2003;50:440–4
15. Ali SZ, Taguchi A, Holtmann B, Kurz A. Effect of supplemental pre-operative fluid on postoperative nausea and vomiting. Anesthesia. 2003;58:780–4
16. Moretti EW, Robertson KM, El-Moalem H, Gan TJ. Intraoperative colloid administration reduces postoperative nausea and vomiting and improves postoperative outcomes compared with crystalloid administration. Anesth Analg. 2003;96:611–7
17. Yogendran S, Asokumar B, Cheng DC, Chung F. A prospective randomized double-blinded study of the effect of intravenous fluid therapy on adverse outcomes on outpatient surgery. Anesth Analg. 1995;80:682–6
18. Hausel J, Nygren J, Thorell A, Lagerkranser M, Ljungqvist O. Randomized clinical trial of the effects of oral preoperative carbohydrates on postoperative nausea and vomiting after laparoscopic cholecystectomy. Br J Surg. 2005;92:415–21
19. Griffenhagen GB, Hawkins LL. Facts and Comparisons: Handbook of Nonprescription Drugs. 6th ed. 1989 Washington, DC American Pharmacists Association:107
20. Lauwick SM, Kaba A, Maweja S, Hamoir EE, Joris JL. Effects of oral preoperative carbohydrate on early postoperative outcome after thyroidectomy. Acta Anaesthesiol Belg. 2009;60:67–73
21. Cook R, Anderson S, Riseborough M, Blogg CE. Intravenous fluid load and recovery: a double-blind comparison in gynaecological patients who had day-case laparoscopy. Anaesthesia. 1990;45:826–30
22. Spielberger CD. Manual for the State-Trait Anxiety Inventory. 1983 Palo Alto, CA Consulting Psychologists Press
23. Dabu-Bondoc S, Vadivelu N, Benson J, Perret D, Kain ZN. Hemispheric synchronized sounds and perioperative analgesic requirements. Anesth Analg. 2010;110:208–10
24. Dabu-Bondoc S, Drummond-Lewis J, Gaal D, McGinn M, Caldwell-Andrews AA, Kain ZN. Hemispheric synchronized sounds and intraoperative anesthetic requirements. Anesth Analg. 2003;97:772–5
25. White PF, Tang J, Hamza MA, Ogunnaike B, Lo M, Wender RH, Naruse R, Sloninsky A, Kariger R, Cunneen S, Khalili T. The use of oral granisetron versus intravenous ondansetron for antiemetic prophylaxis in patients undergoing laparoscopic surgery: the effect on emetic symptoms and quality of recovery. Anesth Analg. 2006;102:1387–93
26. Fujii Y, Itakura M. A prospective, randomized, double-blind, placebo-controlled study to assess the antiemetic effects of midazolam on postoperative nausea and vomiting in women undergoing laparoscopic gynecologic surgery. Clin Ther. 2010;32:1633–7
27. Chen YH, Zhou XH. Interval estimates for the ratio and difference of two lognormal means. Stat Med. 2006;25:4099–113
28. Friedrich J, Adhikari N, Beyene J. The ratio of means method as an alternative to mean differences for analyzing continuous outcome variables in meta-analysis: a simulation study. BMC Med Res Methodol. 2008;8
29. Ledolter J, Dexter F. Analysis of interventions influencing or reducing patient waiting while stratifying by surgical procedure. Anesth Analg. 2011;112:950–7
30. Wachtel R, Dexter F, Epstein R, Ledolter J. Meta-analysis of desflurane and propofol average times and variability in times to extubation and following commands. Can J Anaesth. 2011;58:714–24
31. Garner MJ, Altman DG. Confidence intervals rather than p values: estimation rather than hypothesis testing. BMJ. 1986;292:746–50