Should Noncardiac Chest Pain Be Treated Empirically?:  A Cost-effectiveness Analysis FREE

CHEST PAIN is a common clinical problem. After abdominal pain, it is the most frequent reason for urgent office visits.¹ The prime concern on assessment of patients with chest pain is ruling out coronary artery disease. However, as many as 30% of patients with typical or atypical angina have normal coronary arteries on angiography.² It has been estimated that at least 180 000 cases of noncardiac chest pain (NCCP) are diagnosed yearly, with annual health care costs of $750 million for this group.³ Subsequent investigations of these patients suggest that up to 60% have an esophageal source for the pain, with gastroesophageal reflux disease (GERD) being the most commonly associated abnormality.^3- 5

There are currently no published clinical trials addressing the issue of empirical treatment vs workup and subsequent treatment of patients with NCCP. There are consensus guidelines from the American Gastroenterology Association for the diagnosis of NCCP, but these only review the available investigations and do not recommend a particular sequence or discuss empirical treatment.⁶ Experts in the field have suggested algorithms for management but do not endorse a particular strategy.³ Empirical treatment for GERD (or ulcer disease) vs investigation has been addressed in dyspeptic patients, who may bear some similarities to patients with NCCP. This has been done via clinical trials and through decision analytic models. Unfortunately, these study results conflict and cannot be extrapolated to patients with NCCP.^7- 10 Existing guidelines support empirical treatment for GERD for patients with dyspepsia as well as those with reflux symptoms.^11- 13 However, optimal treatment of patients with NCCP remains undefined.

Given the expense associated with investigating to determine a cause for NCCP, there is reason to believe that empirical treatment might be cost-effective. To address this question we performed a cost-effectiveness analysis comparing empirical treatment with initial upper gastrointestinal (UGI) tract investigation in patients with NCCP.

We developed a decision tree to compare the strategy of empirical treatment of patients who are considered to have a noncardiac source for their chest pain with initial investigation for gastrointestinal disease. All models were constructed and analyzed using DATA software.¹⁴ The assumptions used to construct our decision tree are outlined in Table 1.

Initial decision trees were developed based on extensive literature review and the investigators' clinical expertise. The final tree structure was derived through a Delphi process by a panel of 5 general internists and 3 gastroenterologists.¹⁶ Pilot-testing initial instructions and preliminary trees on 5 general internal medicine fellows at Boston University resulted in clarification of the instructions for the Delphi panel.

Delphi participants were initially asked to choose the most appropriate sequence of investigations and comment on potential test outcomes. They also chose among several treatment options, including trazodone, imipramine, calcium channel blockers, and various GERD treatments.^15,17- 18 With each successive Delphi round the tree structure was revised as consensus evolved. A total of 4 complete rounds was required. Where consensus was not achieved, items were examined as part of the sensitivity analysis. The final decision tree structure is shown in Figure 1.

Probabilities in the model were obtained through an extensive literature review (MEDLINE search and reference review). All NCCP studies and reviews from 1966 through August 31, 1998, were examined. As there are no randomized controlled trials and very few open trials addressing treatment in this patient group, probabilities for treatment response were obtained from review of the GERD literature. Where data were not available, probabilities were derived through the Delphi process. Because many NCCP studies excluded patients with structural lesions on endoscopy or barium studies, comparison was also made with studies in the dyspepsia literature that used an unselected patient population (Table 2).

Cost analysis was done from a third-party payer perspective, considering the direct medical costs of each strategy. For the study arm this included test costs as well as the 8-week treatment course initiated as a result of a positive test result. Treatment arm costs included treatment costs followed by the cost of study and subsequent treatment in the event of initial treatment failure. Investigation costs were derived from Boston Medical Center 1996 charge data; these were converted to costs using cost center–specific cost-to-charge ratios. Provocative test costs were obtained from Cleveland Clinic 1996 charge data, as provocative tests are not performed at the Boston Medical Center. Medication costs were based on the 1996 Red Book average wholesale price (Table 3).

The effectiveness of an initial treatment strategy was compared with the effectiveness of an initial investigation strategy by determining the cost per case of chest pain successfully managed. Based on our literature review we assumed that quality of life and prognosis were not significantly affected by the initial management strategy.^{19- 26,28,47- 48} Since we assumed both arms were equally effective, the least costly strategy is most cost-effective.

One-way sensitivity analyses were performed by varying the base-case probability and cost estimates. Two-way sensitivity analyses were performed for selected variables, including the probability of symptomatic improvement with ranitidine and omeprazole. We also examined the additional costs of test complications and drug toxic effects. Two panel members preferred to empirically treat with a proton pump inhibitor initially for less than 8 weeks, so we analyzed this as a third treatment option. We also examined the cost of including a gastroenterology consultation.

Since previous decision analyses dealing with dyspepsia have found similar costs after 1 year in those treated empirically vs those having initial endoscopy, longer-term costs were also addressed.⁹ Costs for both arms were modeled out to 60 weeks (8 weeks of initial treatment plus 52 weeks of follow-up). There was no cost discounting over the 60 weeks (Figure 2 and Table 4).

In the base-case model, the investigation arm cost $2187 per case treated compared with $849 for the empirical treatment arm. Using one-way sensitivity analyses the model was very robust. The treatment arm was less expensive in all cases. Of note, the treatment arm cost was reduced to $744 if omeprazole, 20 mg/d, was given for 8 weeks as the only option, and increased to $901 if ranitidine was the only option with no second-line treatment. The least expensive strategy was to give the patient ranitidine initially in all cases and then give second-line treatment with omeprazole in the event of ranitidine treatment failure before proceeding to study; the cost was $484. Using two-way sensitivity analysis the study arm became less expensive only if the probability of symptomatic improvement for both treatment options was less than 20%. The cost of the study arm was relatively insensitive to the sequence of tests after endoscopy. Table 5 shows selected one- and two-way sensitivity analyses.

In terms of complications, endoscopy is the only test with any potential for significant morbidity or mortality. The actual risk is negligible, with a 0.005% mortality rate and 0.08% to 0.128% morbidity rate.^47- 48 (These complication rates include groups with sicker, older patients.) Assuming a cost of 1 hospital day at $1000 daily for each endoscopic complication, overall endoscopy costs would increase by only $1.28.

The cumulative probability of an adverse drug reaction was estimated at 5% for both ranitidine and omeprazole based on the 1996 Physicians' Desk Reference. This included diarrhea, abdominal pain, and vomiting. We assumed that a 5% toxic effects rate would result in a 5% probability of discontinuing the medication and proceeding to the investigation arm. This increased the treatment arm cost by 12.5% to $956.

Inclusion of a 2-week treatment option, either as the only option or in addition to the other 8-week treatment options, resulted in no appreciable change in the treatment arm costs. (If an initial 2-week course of omeprazole, 20 mg daily, were given, followed by investigation or a second 2-week course of omeprazole, 40 mg daily, in the case of treatment failure, then the treatment arm would cost $854.) Inclusion of a gastroenterology consultation increased the study arm cost to $2451.

In the long-term model, the treatment arm is still dominant at 60 weeks. The study arm costs $2730 and the treatment arm costs $2270. If only omeprazole were used initially in the treatment arm, the cost would be slightly lower, at $2135. If the only tests performed were UGI series and endoscopy, the study and treatment arms cost $1756 and $1590, respectively.

This study compared initial investigation for a UGI source with empirical treatment for presumed GERD in patients with NCCP. From the base-case 8- to 16-week decision analytic model, empirical treatment was less expensive than initial investigation. The base-case model was very robust. Only when the probability of symptom improvement with both ranitidine and omeprazole was less than 20% was the investigation arm less expensive (Table 5). Neither endoscopic complications nor drug toxic effects were significant factors in increasing model costs.

Beyond 1 year, the empirical treatment remains dominant, although the absolute difference in costs is less than at 16 weeks. This occurs because a patient in the study arm would usually have a negative study result and therefore not receive long-term treatment according to the model, while the chance of an individual in the treatment arm proceeding to study increases with time because of the assumed relatively high rate of relapse or symptom recurrence as estimated from the GERD literature.^46,49- 53

This is the one of the first studies to address the question of optimal management of NCCP. There are comparable studies from the dyspepsia and GERD literature. However, these usually deal only with endoscopy or UGI series vs empirical treatment, while this model includes additional investigations. This model also uses omeprazole, which belongs to the most effective class of antisecretory agents on the market (proton pump inhibitors), as a treatment option. Proton pump inhibitors have not been used in previous dyspepsia studies.

Of the existing decision analytic models in dyspepsia, Silverstein et al⁹ found that the "direct medical charges" of endoscopy vs empirical treatment with ranitidine were comparable at 1 year. Read et al¹⁰ found that empirical treatment costs were lower; however, their model was based on a 6-week period. Of the available models from the GERD literature, Hillman et al⁵⁴ compared empirical treatment using either omeprazole or ranitidine with investigation by endoscopy followed by 24-hour pH monitoring if the results of endoscopy were negative. At 7-month follow-up, omeprazole was the less expensive treatment strategy, with a toss-up between the empirical and investigation arms. (The model differed somewhat in that patients with treatment failure of initial ranitidine therapy did not receive omeprazole, proceeding to surgery instead.) Heudebert et al⁵⁵ modeled a strategy of initial endoscopy or UGI series vs empirical H-blockers or proton pump inhibitors over a 1-year period. Both empirical strategies were less expensive (omeprazole was associated with a higher number of quality-adjusted life years).

Two randomized clinical trials have compared empirical treatment as GERD with initial investigation in dyspeptic patients.^7- 8 (There have been no such trials in NCCP patients.) These trials have produced conflicting results. Goodson et al⁷ compared high-dose antacids with initial UGI series. Treatment was continued until failure or up to 6 months. At 6-month follow-up, average costs for the empirical group were less than half those for the study group. There were no significant differences between the two groups in symptom scores or disability. Bytzer et al⁸ compared empirical ranitidine therapy with initial endoscopy. Endoscopy was performed in the case of initial treatment failure or symptomatic relapse. At 1 year there were no differences in symptoms or quality-of-life measures. However, costs were higher in the empirical arm because of ongoing H-blocker use (and more sick days). Thus, our models are essentially consistent with the dyspepsia and GERD literature. In patients with NCCP, initial costs are clearly less in those treated empirically, while at just over 1 year the difference in costs between treatment and investigation remains but is not as marked.

Recently, in the NCCP literature, Fass et al⁵⁶ addressed the issue of NCCP management in a somewhat different manner. They conducted a randomized controlled crossover trial of omeprazole, 60 mg daily for 7 days, as a diagnostic test for GERD in 37 NCCP patients. A positive response to omeprazole therapy (more than 50% symptom improvement) was considered diagnostic of GERD, and this was compared with endoscopy and 24-hour pH monitoring results. They used decision analysis to compare the costs of an empirical treatment strategy following a positive response to the "omeprazole test" with the costs of initial investigations. In their cost minimization analysis the empirical arm was also less costly compared with initial investigation at 1 year. Their algorithm differs from that used in our empirical arm in that manometry is performed before endoscopy if the omeprazole test result is negative. If the findings of manometry are abnormal, patients receive treatment with diltiazem for a year, even though there is evidence to suggest that patients with motility disorders may still be more likely to have GERD as the cause of their NCCP.⁵⁷ Thus, the true base-case cost estimate of the empirical arm may be somewhat higher, although still less than the investigation arm, as in our study.

Although clinical trials are the optimal method to answer clinical questions, decision analysis is a useful tool when such trials are lacking. One of the strengths of our decision analysis lies in the fact that our algorithm was generated using a Delphi panel of experts who regularly treat patients with NCCP. Although literature is emerging suggesting use of omeprazole as first-line medical treatment in GERD and possibly NCCP by extrapolation, at the time of our Delphi panel this was not the consensus.^55,58 However, our sensitivity analyses incorporated differences in opinion with respect to both treatment and investigation, including the use of omeprazole alone.

Our model, like all decision analyses, is limited by the validity of the underlying assumptions and the data quality. Our model hinges on the assumption that abnormalities of the esophagus, especially GERD, constitute the most common cause of NCCP. This is supported by case series that report a high incidence of esophageal abnormalities, most frequently GERD, on investigation, and small randomized trials of patients with NCCP and GERD whose condition improved with reflux treatment.^{4- 5,20,59- 61}

Although it is increasingly believed that GERD is the most common cause of NCCP, other possible nongastrointestinal causes cited in the literature include microvascular angina, psychiatric disorders such as panic attacks, and mitral valve prolapse.^62- 65 It is likely that these are much less common causes of NCCP than GERD. Moreover, studies suggest that there may be considerable overlap between these conditions and esophageal-related chest pain.^34,66- 67

Other gastrointestinal causes of NCCP cited include gallbladder disease.^3,68 However, in the few NCCP series that have reported any cases of gallbladder disease, there was no chest pain improvement in patients who underwent cholecystectomy.⁶⁹ Hiatal hernia, gastritis and duodenitis were not considered significant test findings in our study, and existing evidence does not support them as causes of NCCP.^9,70- 71

In our model, the prototype patient was female, aged 45 to 55 years, with typical angina-like pain. Based on the literature, there is little reason to believe, however, that the model would be altered if the patient were male or had atypical chest pain. We used a woman in our model because most of the case series reviewed had a slight female preponderance. The age of our patient is based on the average age reported in most series. However, ages ranged from as low as 18 to as high as 76 years, so our results should be applicable to a wider age group.^19- 26,28

Despite the excellent survival prognosis of patients with NCCP, concerns remain about missing serious structural UGI tract pathologic conditions.^69,72- 73 However, there were no cases of esophageal or gastric cancer in any reviewed series.^19- 26,28

Concerns also exist regarding ongoing morbidity associated with the patient not knowing why he or she has chest pain and the cost associated with treating a patient in the absence of a firm diagnosis. Follow-up studies in NCCP patients have consistently shown decreased rates of hospitalization.⁷⁴ However, these patients often continue to have activity limitations, ongoing chest pain, time lost from work, and medication use.^74- 78 Proponents of gastrointestinal investigation argue that patients have less morbidity if they know the cause of their pain. The 3 follow-up studies specifically addressing this issue include patients referred to a gastrointestinal laboratory, who presumably are already more concerned about their chest pain. They have found conflicting results, with 2 of 3 studies showing no benefit from the investigations.^38- 39,79

In summary, empirical treatment as GERD is more cost-effective than initial investigation up to at least 60 weeks. Most of the savings occur early on, with the difference still existing but less marked at over a year. However, presumably, not all patients will require long-term medication, and some may have resolution of symptoms with no recurrence after an initial 8-week course of antisecretory agents, which would further minimize costs. Confirmation of the optimal empirical strategy awaits further clinical trials.

Accepted for publication July 12, 1999.

Presented as a poster at the 20th Meeting of the Society of General Internal Medicine, Washington, DC, May 1-3, 1997.

The authors thank Dan Berlowitz, MD, Graeme Fincke, MD, Kenneth Freedberg, MD, Daniel Friedlander, MD, Mark Moskowitz, MD, Raj Goyal, MD, Jay Orlander, MD, and John Saltzman, MD, for their participation in the Delphi panel. In addition, we thank Drs Berlowitz, Fincke, and Moskowitz, as well as Eli Schimmel, MD, for their editorial assistance.

Corresponding author: Ann M. Borzecki, MD, MPH, Room 440, Bethune Building, QEII HSC, 1278 Tower Rd, Halifax, Nova Scotia, Canada B3H 2Y9 (e-mail: borzecki@is.dal.ca).