PureInsight | March 11, 2002
Purpose: To derive a clinical decision support model in keeping with Truthfulness-Compassion-Forbearance for helping the physician and his patient determine the best treatment modality, laryngectomy vs. laryngeal preservation with combined chemo- and radiotherapy, for patients with stage II and III squamous cell carcinoma of the larynx with or without nodal involvement (T2/T3, N0/N+).
Method: A Markov model was formulated to aide in the decision analysis using the software package TreeAge (student version). Markov states were determined based on the rehabilitative modalities after total laryngectomy, death, recurrences, and disease-free survival after chemo/-radiotherapy with or without late effects of the therapy. Values for survival and recurrence rates after the therapeutic modalities were drawn from a cohort of patients treated at MDACC with the same characteristics as those eligible for the experimental protocol.
Results: Since the model draws on the experience of the entire cohort of patients with similar disease and complex calculations of utility and disutility are done without caretaker bias, this decision support system assimilates to Truthfulness. Since the model calculates the most likely results of the various treatments and rehabilitative modes and the patient's subjective responses to those states, it also is a more compassionate way of helping the patient and physician decide upon the course of therapy. The therapeutic and rehabilitative regimens chosen with the assistance of this model will require large measures of forbearance or endurance on the part of all participants if they are to succeed. Quality-adjusted survival for laryngectomy +/- neck dissection was higher (4.47) than that for the laryngeal preservation procedures (3.94), but the analysis was very sensitive to quality of life parameters, indicating that it is very important to take into account the patient's reactions to the various possible outcomes.
Introduction
The recommendations for medical care for patients who seek the advice of physicians should reflect the fundamental principle of Truthfulness-Compassion-Forbearance. Life or death decisions concerning treatment of serious diseases can be influenced by physician bias or incomplete knowledge. For example, as treatment for a patient with a certain type of cancer, a surgeon may be likely to recommend surgery as the definitive treatment method, while a radiation oncologist might recommend radiotherapy, and a medical oncologist might recommend chemotherapy, all based on their own experiences. Such biased judgments may well fail to reflect the truth of the situation and thus be less than compassionate with respect to the patient. Computer based clinical decision support systems can perform complicated calculations involving many variables and their mutual interactions which depend on the specifics of the patient's clinical and demographic states and represent the entire medical community's experience with these parameters. These systems thus can help a physician or a patient truly evaluate the relative efficacies of various treatment modalities in a given situation. But analysis using these models reflects responses of large populations of patients, not the individual under consideration. Furthermore, the objective data that are used as inputs for these analyses do not include subjective qualities of the patient that may well have a very large bearing on the clinical outcome and, very importantly, on the patient's perception of that outcome. For some patients, for example, any disfigurement or loss of function from surgical procedures would be a devastating blow to their psyches while others are more likely to develop coping strategies for dealing with immediate loss but would prefer to avoid irremediable side effects of chemo- and radiotherapy that can take years to develop. For compassionate reasons, the patient's perceptions of his ability to cope under these various possible scenarios should be taken into account when he and his physician decide upon his course of therapy.
Squamous cell carcinoma is the most common type of laryngeal cancer. The treatment protocol ranges from radiation therapy, chemotherapy, and surgery to various combinations of these options (3). Treatment modality for most stages I and II laryngeal cancer is radiotherapy (1, 2). Patients with stage III and IV are usually treated with surgery (total laryngectomy) with or without radiotherapy (2) but many patients find the prospect of the loss of their larynx truly daunting. While radiotherapy alone is not effective in treating higher stage laryngeal carcinomas, its combination with chemotherapeutic measures does offer promise of curative effects. This study was undertaken to begin to develop a decision support model with Truthfulness-Compassion-Forbearance in order to help the patient and his physician assess whether, on the basis of recurrence and survival rates of patients with similar disease so treated, in conjunction with quality of life issues, these two basic approaches to treatment for T2-3N0-3 squamous cell carcinoma of the larynx were comparable in terms of quality adjusted survival or whether one approach was better than the other for this particular patient.
Case
Consider a patient, 57 year-old white male, a trial lawyer, who presents with a lump in the neck and a history of smoking and alcohol use. The patient was previously untreated. The diagnosis could be either a stage II or III laryngeal cancer with nodal involvement (T2/T3, N+).
Treatment options presented to the patient include: surgical treatment, laryngectomy with (in this case) or without neck dissection, or larynx preservation treatment, combined chemo- and radiotherapy.
The patient would prefer to keep his voice, but would also prefer to keep his life. He has two questions: 1) Does laryngeal preservation therapy offer roughly comparable cure rates compared to laryngectomy? And 2) Would recent advances in Speech Therapy for laryngectomees and the possibility of debilitating late effects from the chemoradiotherapy change the apparent desirability of the two options?
For the purpose of building the model in this very preliminary study, the data used is from a cohort of similar patients treated at the U. T. M. D. Anderson Cancer Center from 1985 to mid-1999.
Model
Given the treatment options for T2/T3, N+ laryngeal cancer, the decision to be made is whether to opt for surgery (Laryngectomy +/- Neck Dissection) over laryngeal preservation (Chemotherapy/ Radiotherapy). Conservative treatment with chemotherapy/ radiotherapy is the treatment opted for by most patients (3).
The root of the tree is a decision node for the T2/T3, N+ laryngeal cancer and has two branches:
Laryngectomy +/- neck dissection and Laryngeal preservation. If the decision to opt for surgery is followed there is a chance of success or failure of the surgery. Failure results in mortality whereas if the larynectomy is successful the patient survives. The operative mortality from surgery (pOpMort1) was 1%. Operative mortality from salvage surgery for radiation failure reported was 7% (3). In the event of survival, the patient is in either one of the Markov states (4).
Markov States:
Following successful surgery five Markov states were established based on rehabilitative speech modalities:
No talk
Electrolarynx
TracheoEsophageal speech
Recurrence and Death
The absorbing states being recurrence of the cancer and finally death. (4)
Following complete response to chemotherapy/ radiotherapy the Markov states were:
No Evidence Cancer
No Evidence Cancer w/ Late Effect
Recurrence and Death
Assumptions for the Markov Model:
Several assumptions were made in the formulation of this model:
1. Failure of the larynectomy results in the death of the patient.
2. The Markov states were determined with the assumption that if the patient survives the surgery he would be in any one of these five states.
3. For simplicity of the model recurrence of the cancer after surgery was assumed to be synonymous to being dead because salvage surgery would not be performed.
4. The various rehabilitative states were determined based on the assumption that there was no evidence of cancer at the time.
5. Esophageal speech was not included as a separate state but was combined with tracheoEsophageal speech.
6. Incomplete preservative therapy was a result of discontinuing therapy due to it side effects.
7. Quality of life is lower after the complete response of the tumor due to late effects from chemotherapy/ radiotherapy.
The Markov-cycle tree begins with a Markov node with the five Markov states mentioned; No talk, Electrolarynx, TracheoEsophageal speech, Recurrence and Death.
Initial Probabilities:
All patients after laryngectomy start off with No talk state, assigned a probability of 1. In the case of laryngeal preservation with complete response to chemotherapy/ radiotherapy, all patients begin with No Evidence Cancer state, probability assigned was 1.
A subtree follows each Markov node for a particular cycle. Patients start off in the No talk state (chance node) with a chance of dying from various causes, hence the branch Die, with terminal node Dead. If the patient does not die, Recurrence of the cancer can occur. Recurrence models a chance node for dying (pDie1), which is terminal or Alive with recurrence with terminates in the Recurrence state. If the patient does not die or recurrence does not occur, there is a chance of staying in the same state of health (Stay No Talk), which terminates in the No Talk state. The patient could progress form no talk to adopting either one of the rehabilitative modalities Electrolarynx or TracheoEsophageal speech, which would terminate in each of the respective Markov states.
Similarly the remaining subtrees were constructed with the exception that once in either of the other two states Electrolarynx or TracheoEsophageal the patients will not make a transition to the No talk state.
The subtree following complete response to conservative treatment with chemotherapy/ radiotherapy, the Markov cycle tree includes four Markov states No Evidence Cancer w/ Late Effect, No Evidence Cancer, Recurrence and Death
Patients start off in the No Evidence Cancer state (chance node) with a chance of dying from various causes, hence the branch Die, with terminal node dead. If the patient does not die, Recurrence of the cancer can occur. Recurrence models a chance node for dying (pDie1), which is terminal or Alive with recurrence with terminates in the Recurrence state. If the patient does not die or recurrence does not occur, there is a chance of staying in the same state of health (Stay No Evidence Cancer), which terminates in the No Evidence Cancer. The patient can also develop late effects, which terminates in No Evidence Cancer w/ Late Effect.
Cycle Length
For the purposes of description of this model a yearly cycle was used for a period of 10 years.
Assigned Probabilities
Different baseline mortalities were assigned for each cycle given the age of the patient (57 years, white male).
Table 1. tASR (5)
Assigned Utilities:
Incremental utilities assigned for each of the Markov states were based on the quality of life for each rehabilitative Markov state following surgery.
No Talk state, qS, quality of life from being silent was assigned a value of 0.5. An initial reward was adjusted to 0.5*qS or the incremental utility.
Electrolarynx state: qEL, quality of life from the use of a electrolarynx was assigned as value of 0.7, with initial reward adjustments of 0.5*qEL
TracheoEsophageal qTES, quality of life from the use of a tracheoesophageal speech was assigned a value of 0.85, with an initial reward adjustment of 0.5*qTES.
Recurrence and Death were assigned incremental and initial rewards of zero.
Similarly, Markov states of No Evidence Cancer w/Late Effects and No Evidence Cancer were assigned utilities accounting for the quality of life from the occurrence of late effects, qLE value assigned was 0.85, and adjusted for initial rewards.
Figure 1 is a diagram of the structure of the decision tree with the default values obtained from the medical community's experience with similar patients inserted. The specific patient's perceptions of future quality of life issues would then be derived from one or more interviews and responses to items on a survey form. An example of the survey form used in this study is attached as an appendix.
Results and Discussion:
1QOL Silence
2QOL Electrolarynx
3QOL Tracheo-esophageal Speech
4QOL Late Effects of Chemoradiation
5Quality-adjusted Survival in Years
Table 2. Effects of patient estimations of future quality of life on quality-adjusted survival after laryngectomy or chemoradiation for stage II/III squamous cell carcinoma of the larynx.
Sensitivity analysis determined that the choice of preferred modality varied with variations in probabilities of side effects, and incomplete responses, as well as in response to changes in quality of life disutilities. It would have been interesting to measure effects of changes in probabilities of death and recurrences and late effects after the various treatments but, since these values were in tables based on real data, they could not be varied for "what if?" analysis.
If the QOL values listed above are set to 1.0, the "quality-adjusted" survival is higher for laryngectomy than for laryngeal preservation therapy, probably because of the lower recurrence rate for immediate laryngectomy, as opposed to chemoradiation or laryngectomy after failure of chemoradiation, and the lower survival rate for patients who receive salvage laryngectomy.
Respondent estimations of their future qualities of life in the various rehabilitative states after laryngectomy and with late effects after chemoradiation, compared to their estimated qualities of life one year ago, are reflected in the quality-adjusted survival values calculated by this model for the two treatment modalities under consideration. Therefore, after modifications suggested by this and further studies, the model might have some utility as a clinical decision support tool.
Respondent estimations of their future qualities of life after the various treatments varied widely between individuals. This could be reflective of individual propensities but also could be indicative of different levels of familiarity with the states. If a survey procedure like this one were to be implemented in a clinical setting, it would undoubtedly be beneficial to expose the patients to training sessions about the various states before surveying their preferences.
The use of a decision support modeling system like this one advances the application of Truthfulness-Compassion-Forbearance to helping the physician and his patient determine which course of therapy and rehabilitative procedures are most suitable for that particular patient. Because it utilizes the experiences of the entire cohort of patients with similar disease who have undergone these treatment and rehabilitation modalities and then calculates the resultant quality adjusted survival values without caretaker bias, it more closely reports truthfully the past experience in the situation. Because it gives a true picture of past outcomes and carefully takes into account the patient's perceptions of the quality of life he would experience given the various outcomes, it is a compassionate way to arrive at the best regimen for a given patient. The use of such a decision support system would be especially compassionate if it is coupled with educational materials about the possible late side effects of the various treatments and full explanations of the rehabilitative options that are now available, along with psychological and spiritual counseling. Part of this counseling would entail emphasis on the forbearance or endurance qualities the patient and his caretakers must exhibit for the best possible results.
References:
1. Parsons JT, Mendenhall WM, Stringer SP, Cassisi NJ, Million RR
Salvage surgery following radiation failure in squamous cell carcinoma of the
supraglottic larynx. Int J Radiat Oncol Biol Phys 1995 Jun 15;32(3):605-9
2. Yuen AP, Ho CM, Wei WI, Lam LK Prognosis of recurrent laryngeal carcinoma after laryngectomy. Head Neck 1995 Nov-Dec;17(6):526-30
3. Yuen AP, Wei WI, Ho CM Results of surgical salvage for radiation failures of laryngeal carcinoma. Otolaryngology Head Neck Surg 1995 Mar; 112(3): 405-9
4. Sonnenberg FA, Beck JR, Markov Models in Medical Decision Making, Med Decis Making 1993;13:332-338
5. National Vital Statistics Report, U.S. Life Tables, National Center for Health Statistics, Dec 13, 1997, 47(28)
