Medical or Surgical Therapy for Erosive Reflux Esophagitis: Cost-Utility Analysis Using a Markov Model

Joseph Romagnuolo, MD, FRCP(C), MSc, (Epid); Michael A. Meier, MD, FRCS(C); Daniel C. Sadowski, MD, FRCP(C)


Annals of Surgery. 2002;236(2) 

In This Article


A two-stage Markov model was constructed using the DATA (TreeAge, Boston, MA) software package.[30] Our base case was a 45-year-old man with endoscopically proven grade II-IV erosive esophagitis, refractory to H2-blockers. We chose a time horizon of 5 years because of currently available data on relapse rates.[31] Our simulation considered two treatment options: medical therapy with omeprazole for endoscopic healing and maintenance versus surgery using LNF.

The Markov model of prognosis is one example of an alternative to standard decision analytical techniques wherein transitions between health and various disease states are modeled over time.[32] Instead of assuming, for example, a one-time probability of therapeutic failure as in a standard decision analysis, the model instead estimates a certain probability for failure during each time cycle.

One intrinsic limitation to the Markov model is the lack of memory.[32] The model assumes that the previous experience of the person in the model does not influence the probabilities of subsequent events (i.e., transition probabilities are independent of one another). In our analysis, a certain degree of memory was required for the medical arm of the model, as the maintenance dose was dependent on the healing dose required. Thus, a two-stage model was constructed, which included the creation of five separate Markov chains stemming from the five regimens required for successful healing (described below). There is evidence that the initial grade of esophagitis does not influence outcome significantly after LNF, and so modeling of memory for the surgical arm was not performed.[33]

Figure 1 illustrates the mathematical relationships of the health states used in our Markov model. Transitions were allowed at the end of each 3-month cycle. Rates derived from the medical literature were converted to transition probabilities according to previously published methods (using the formula P(t) = 1 - e-rt, where P(t) is the probability of an event during the time cycle, r is the rate, and t is the duration of the cycle expressed in the same time units used in the rate).[30] Rates were varied using interpolation of an approximated hazard function derived from the literature for transitions for which a constant hazards assumption was not appropriate over the 5-year time period (symptomatic relapse and transition between dosages of omeprazole, postoperative dysphagia).[25,31] Age- and gender-matched population-based mortality rates (life expectancy 31.8 years and 37.0 years for 45-year-old men and women, respectively) came from published Alberta vital statistics.[34] At the completion of each cycle, costs and utilities were accrued.

Markov states and possible transitions modeled for the maintenance phases of the two treatment options: (A) omeprazole and (B) laparoscopic Nissen fundoplication.

Utilities were used as an estimation of quality of life, ranging from an asymptomatic cycle, which accrues 0.25 quality-adjusted life-years (QALYs; i.e., three "healthy" months), and death, which is assigned a utility of 0. Disutilities, which were subtracted from this ideal utility, were assigned to disease states such as recurrent reflux symptoms or postoperative recovery ( Table 1 ). A recent Canadian study used standard gamble and time trade-off techniques in symptomatic reflux patients to estimate the disutility associated with chronic GERD.[35] The investigators found that reflux symptoms were associated with a utility of 0.90 to 0.97, depending on the technique employed. Our model used the mean of these two figures. This disutility was applied for 1 month for each patient who relapsed during a cycle. For those individuals in the medical arm, the disutility associated with chronic omeprazole ingestion was assumed to be similar to that of other chronic medical therapy, such as lipid-lowering agents (0.01).[36] Surgery was assumed to have a disutility of 0.5 for the immediate 2-week postoperative period. For those surgical patients who required dilatation or foreign body removal, the assigned disutility was 0.20 for 1 week. All utilities were discounted at 3% per annum in the maintenance phase.

For the cost analysis, we took the perspective of a Canadian provincial (Alberta) health ministry. Discounted direct costs (in Canadian dollars) were estimated from local costs incurred by the hospital, drug formulary, or provincial health ministry ( Table 2 ). Selected costs were estimated from charges billed to nonresidents. The cost of LNF was estimated from a local costing study carried out at the Grey Nuns Hospital in Edmonton, Canada. Mean length of stay (1.2 days), consumables, and other hospital costs for LNF were calculated by reviewing all such surgeries performed during a 3-month period. Ranges in hospital stay reported in the recent literature were considered in sensitivity analyses. Per-diem costs were estimated from charges billed to non-Alberta residents (Grey Nuns Hospital). Physicians' fees were derived from the fee schedule of the Alberta Health Care Insurance Plan. For the purposes of the analysis, the financial cost of death was assumed to be zero. All costs were discounted at 3% per annum in the maintenance phase.

The medical arm consisted of an initial healing phase followed by maintenance therapy, both using the PPI omeprazole. In the healing phase, patients were assigned to one of five treatment arms, each one representing the different dose and/or duration of therapy required to accomplish successful endoscopic healing. We based the proportion of patients assigned to each of the five healing regimens primarily on probabilities derived from the work of Klinkenberg-Knol et al, who published detailed healing and maintenance data for patients with refractory erosive (grade II-IV) esophagitis, using endoscopic criteria for healing success and relapse (see Table 1 ).[31] In this cohort, omeprazole at 40 mg daily healed 64% of patients after 1 month, with an additional 18% healing after 2 months and 13% after 3 months. Of the remainder, 4% healed after 4 months, and 1% required 4 months of therapy plus an additional month of omeprazole at 60 mg/day. These observations have been confirmed by other studies.[37,38,39,40,41,42]

After healing, patients in the medical arm entered the maintenance phase at omeprazole 20 mg/day. Patients requiring more than 4 months of therapy or more than 60 mg/day omeprazole to achieve healing received 40 mg/day omeprazole in the maintenance phase. If patients relapsed, their maintenance dose was escalated by 20-mg/day increments to a maximum of 60 mg/day. Probabilities for relapse were derived from the data of Klinkenberg-Knol et al.[31] The following daily doses were required to maintain remission after 1 year of follow-up: 20 mg for 65%, 40 mg for 32%, and 60 mg for 3% of patients. Other investigators have observed similar relapse rates.[43,44,45,46,47] We assumed that individuals in the maintenance group requiring 60 mg/day omeprazole would opt for LNF at a rate of 20% annually. An initial upper endoscopy was required not only for the medical healing phase but also if the patient required 60 mg omeprazole at any time during the analysis. For all maintenance arms, physician visits were required only for symptomatic patients, with a maximum frequency of every 3 months.

In the surgical arm of the model, the healing phase was considered to be the surgery itself. If one survived the surgery, one entered the Markov model for the maintenance phase. In a study published in 1997, Perdikis et al reviewed the current LNF literature, including 2,453 patients and 24 studies, with up to 3 years (median not available) of follow-up.[25] Table 3 summarizes the literature from 1996 until the present for comparison. In Perdikis et al's study, the perioperative mortality rate for LNF was found to be 0.2%. It appears from review of more recent literature that even this low rate may be an overestimation.[17,23,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64] Urgent laparotomy was required for complications (perforation, bleeding, perforation, pneumothorax, splenectomy) in 1.5%, while 5.8% required conversion to an open approach (for reasons of exposure, perforation, bleeding, CO2 retention, or pneumothorax). In comparison, more recent publications report an average conversion rate of 3.3%.[17,23,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64] During follow-up, dilatation was required for 3.5% of patients due to dysphagia or gas-bloat type symptoms, and 0.5% required endoscopy for food impaction. Intermediate to long-term dysphagia rates in more recent reports are not markedly different, with a weighted mean of 4.8%.[17,23,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64] Repeat surgery was required in 1.6%, either for dysphagia/gas-bloat or for recurrent reflux symptoms. In the review by Perdikis et al,[25] 3.4% of patients continued to require antisecretory medications for persisting reflux symptoms.

Based on the above data, we assumed that among patients who developed recurrent symptoms following LNF, one third would elect to have surgery redone (via the open approach), and the remainder would continue on 40 mg/day omeprazole for the duration of the study. Although four small studies have now shown that laparoscopic re-do procedures may be feasible, this option was not modeled, as larger studies were felt to be needed in this area.[65,66,67,68] For the open approach, a mortality rate of 1.4% was used.[69,70,71,72] Repeat surgical failures were maintained on 40 mg/day omeprazole. If 40 mg/day failed, the dose was increased to 60 mg/day ("Fail Med-Surgery" in Fig. 1).

In the surgical arm, all patients underwent endoscopy not only at presentation but also if 60 mg/day omeprazole was required at any time during the analysis, as in the medical arm. Endoscopy was also carried out if the patient required dilatation or foreign body removal, using the rate estimated by Perdikis et al[25] (see Table 2 ). A rate of dilatation, after foreign body removal, of 70% was used.[73]

Sensitivity analyses were performed to test the robustness of the model with respect to the assumed parameter values. Clinically relevant ranges were tested using available literature (see Table 1 ).

In a Monte Carlo simulation, each patient passes through the model from beginning to end (5 years), with transitions at each cycle decided by a random number generator and the probabilities associated with that transition.[32] The purpose of the simulation is to arrive at an estimate of the precision of the outcomes of interest (costs and utilities). For each arm, 10,000 patients were simulated, yielding estimates of the mean costs and utilities for each strategy as well as a standard deviation and 95% interpercentile ranges for each parameter.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: