TY - JOUR
T1 - Inverse decision theory with applications to screening and diagnosis of cervical intraepithelial neoplasia
AU - Davies, Kalatu R.
AU - Cox, Dennis D.
AU - Swartz, Richard J.
AU - Cantor, Scott B.
AU - Follen, Michele
N1 - Funding Information:
The data used in our analyses are part of an ongoing program project, “Optical Technologies for Cervical Neoplasias,” funded by the National Cancer Institute. This project is further described in Ref. [11] . It includes subjects from a screening population, patients who coming in for a Pap Smear, and a diagnostic population, patients who have had a positive Pap Smear and have been referred for a colposcopy. Each data point consists of three observations (Disease State, Pap Smear, Colposcopy) for each subject. Disease state is determined by the (consensus) biopsy which has had at least two readings (a third reading if there was disagreement in the first two). Negative and positive results from the Pap smear and colposcopy will be denoted 0 and 1, respectively. Disease state for negative, low grade, and high grade are denoted as 0, 1, and 2, respectively. The screening and diagnostic populations will be denoted as s and d, respectively. The data are summarized in Table 6 .
PY - 2007/10
Y1 - 2007/10
N2 - Objective.: Medical decision makers would like to use decision theory to determine optimal treatment strategies for patients, but this requires priors, likelihoods, and losses. It can be very difficult to specify a loss or utility function in a medical setting, especially when considering both patient health outcomes and economic costs. These issues led to the development of Inverse Decision Theory (IDT), which involves determining the set of losses under which a given decision rule is optimal. Methods.: We apply IDT to the current standard of care for the diagnosis and treatment of precancerous lesions to the cervix, using a Bayesian approach to estimate the probabilities associated with diagnostic tests and make inferences about the region of optimality. There are two ways in which Inverse Decision Theory can be useful: (i) if the decision rule of interest is optimal, then we obtain information about the losses for the optimal treatment strategy, and (ii) if the decision rule of interest is not optimal, then we characterize the losses under which it would be optimal, and assess whether or not it contains reasonable values of the losses. Results.: This paper introduces important clinical results: in particular, we find that the current standard of care for cervical precancer is probably not optimal, and a new decision rule which requires a confirmatory biopsy for all patients with a positive Pap smear test result is better. Conclusion.: We have developed a very general and flexible approach for evaluating treatment strategies that could prove useful in a variety of medical applications.
AB - Objective.: Medical decision makers would like to use decision theory to determine optimal treatment strategies for patients, but this requires priors, likelihoods, and losses. It can be very difficult to specify a loss or utility function in a medical setting, especially when considering both patient health outcomes and economic costs. These issues led to the development of Inverse Decision Theory (IDT), which involves determining the set of losses under which a given decision rule is optimal. Methods.: We apply IDT to the current standard of care for the diagnosis and treatment of precancerous lesions to the cervix, using a Bayesian approach to estimate the probabilities associated with diagnostic tests and make inferences about the region of optimality. There are two ways in which Inverse Decision Theory can be useful: (i) if the decision rule of interest is optimal, then we obtain information about the losses for the optimal treatment strategy, and (ii) if the decision rule of interest is not optimal, then we characterize the losses under which it would be optimal, and assess whether or not it contains reasonable values of the losses. Results.: This paper introduces important clinical results: in particular, we find that the current standard of care for cervical precancer is probably not optimal, and a new decision rule which requires a confirmatory biopsy for all patients with a positive Pap smear test result is better. Conclusion.: We have developed a very general and flexible approach for evaluating treatment strategies that could prove useful in a variety of medical applications.
KW - Bayesian decision theory
KW - Cervical intraepithelial neoplasia
KW - Medical decision making
KW - Squamous intraepithelial lesions
UR - http://www.scopus.com/inward/record.url?scp=34748865009&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=34748865009&partnerID=8YFLogxK
U2 - 10.1016/j.ygyno.2007.07.053
DO - 10.1016/j.ygyno.2007.07.053
M3 - Article
C2 - 17850855
AN - SCOPUS:34748865009
SN - 0090-8258
VL - 107
SP - S187-S195
JO - Gynecologic oncology
JF - Gynecologic oncology
IS - 1 SUPPL.
ER -