Helping patients make choices about breast reconstruction: A decision analysis approach

Abstract

Decision analysis can help breast reconstruction patients and their surgeons to methodically evaluate clinical alternatives and make hard decisions. The purpose of this paper is to help plastic surgeons guide patients in making decisions though a case study in breast reconstruction. By making good decisions, patient outcomes may be improved. This paper aims to illustrate decision analysis techniques from the patient perspective with an emphasis on her values and preferences. We introduce normative decision-making through a fictional breast reconstruction patient and systematically build the decision basis to help her make a good decision. We broadly identify alternatives of breast reconstruction, propose types of outcomes that the patient should consider, discuss sources of probabilistic information and outcome values, and demonstrate how to make a good decision. The concepts presented here may be extended to other shared decision-making problems in plastic and reconstructive surgery.

In addition, we discuss how sensitivity analysis may test the robustness of the decision and how to evaluate the quality of decisions. We also present tools to help implement these concepts in practice. Finally, we examine limitations that hamper adoption of patient decision analysis in reconstructive surgery and healthcare in general. In particular, we emphasize the need for routine collection of quality of life information, out-of-pocket expense, and recovery time.

I. Introduction

When making decisions about breast reconstruction, it is the patient who must live with the consequences. Therefore, patients should have an active role in the decision-making process and should ultimately be the one to make a choice. However, few people have formal training in making decisions. A patient may be a candidate for multiple medically appropriate methods of breast reconstruction. She may be fraught with uncertainty, which is characteristic of difficult decisions. In addition, human beings are heavily influenced by emotions and cognitive biases which compromise their ability to make fully rational hard decisions [1, 2]. In particular, the “indecisive decision-makers,” those that vacillate between indecision, hesitation, decidedness, and doubt, require the most help [3]. However, the question of how to decide on the proper course of action can readily be conceptualized with the application of decision analysis.

Decision analysis is an exhaustive, iterative process that involves identifying alternatives, obtaining information about the uncertainty of outcomes, and clarifying preferences and values [4–6]. Previously, Sears and Chung provided a general tutorial in decision analysis for physician decision-making [6]. The purpose of this paper is to present shared clinical decision analysis to plastic surgeons though a case study in breast reconstruction. The concepts presented here may prove useful for physicians and patients in efficient and effective decision-making. Decision analysis can be an important tool for patients and their plastic surgeons when making shared, clinical decisions. The goal of this analysis is to make a good decision and differentiate decisions from outcomes.

Decision analysis is applied routinely in industry, particularly in the energy sector, taught extensively in graduate business administration education, helpful in physician-oriented clinical decision-making [6], and even useful in the defense of physicians in medical negligence malpractice suits [7]. We believe that it will be valuable in breast reconstruction decision-making and should be adopted for this purpose.

II. Decisions and Outcomes Are Different

We must clarify the difference between decisions and outcomes. Decisions are not outcomes nor are outcomes, decisions. Outcomes result from decisions. A decision is an action where after having been made the patient will find it difficult to return to her prior state [8]. In making a decision (e.g. whether or not to have reconstructive breast surgery after mastectomy), the patient seeks a “good” outcome (e.g. satisfaction with her breasts). However, outcomes are never completely within her control (e.g. she may ultimately be dissatisfied with the result because of a failed free flap). Thus, even if she makes a “good” decision, she can still experience a “bad” outcome. The only thing she controls is her decision. A good decision is one that takes into account her preferences and the uncertainties inherent in reconstructive surgery (e.g. probability of a failed flap). A good outcome is one that she prefers.

III. Normative Breast Reconstruction Decision-Making

We begin by identifying the decision-maker as the breast cancer survivor who has had mastectomy or is considering mastectomy and is committed to action. Indeed, patients who make their own decisions are more satisfied with their care and possess better quality of life [9–11]. The decision-making is normative in that we want to help the patient identify what she should do [1].

Any decision scenario can be described with three components constituting the decision basis: (1) the available alternatives (i.e. options, what we can do), (2) outcome probability information (what we know), and (3) patient preference probabilities [6, 8, 12–15] (what she likes). When these components are defined, evaluation with reasonable and logical rules identifies the best alternative. These five axioms are commonly given as: (1) possibilities characterize distinctions of alternatives, (2) alternatives may be ordered from best to worst, (3) indifference between equivalent alternatives, (4) free substitution of equivalent alternatives, and (5) preference for options with a higher probability of a favorable outcome given that the possibilities are otherwise the same [5, 6, 16]. We assume that the patient agrees with these five rules. The goal is to use these rules and decision basis to achieve clarity of action.

The patient’s plastic surgeon is a good source for identifying available alternatives of breast reconstruction, as well as the possibilities that characterize them. Decision aids designed for patients also may provide some of this knowledge [17–19]. The patient decides on an alternative; however, the outcomes that follow from that decision alternative are not deterministic. For instance, if the exact same procedure could be performed on the same patient multiple times, the same outcome would not be observed every time.

Probability information must be associated with all the possibilities. Traditionally, the source of information for the probabilities is the physician [20–22]. Individual plastic surgeons may be able to estimate probabilities with some reliability; however, we have shown that a group of surgeons provides increased reliability [23]. This information can also be obtained from data and published literature.

Finally, we must consider the patient’s preferences for possible outcomes. Breast reconstruction patients have identified informational needs that included the length of the process, achievable aesthetic outcomes, sensations of the reconstructed breast and donor site, recovery time, details of post-operative pain and discomfort, and possible complications [24–26]. Only the patient knows her preferences and thus must be consulted for hers. Clarifying preferences is not necessarily straightforward; however, we will present one method of obtaining them.

IV. Patient Values and Preferences: Weights and u-Values

Patient decision-making should be focused on her values [27]. Comprehension and knowledge retention of breast reconstruction facts may be low [26]. However, what are essential to her are the outcomes. Ideally, a patient’s preferences for outcomes should be assessed over a period of time when she is calm (i.e., like preparing an advanced directive). Patients should only concern themselves with what they like and, in this regard, they can be very clear. Breast reconstruction itself is merely a means by which to obtain some desired end (i.e. a reasonable facsimile of her absent breast and a better life). And these ends may be measured.

A breast reconstruction outcome measure value is expressed as an expected value or e-value (e.g., out-of-pocket cost may be expressed as an e-value in dollars). We can express a patient’s preference for an e-value with a preference probability or u-value, also known as a utility that is bound between zero (worst life situation) and one (best life situation). The corresponding u-value of a given e-value may differ between individuals because individuals value things differently. An outcome, therefore, is defined as a set of measures and concomitant e- or u-values. There are two primary considerations of patient preference: (1) measure preference and (2) risk preference.

The patient may place more importance, or weight, on certain measures than others (e.g., a patient with limited financial resources may weigh monetary expense more highly than the amount of time lost to reconstruction). A robust method for obtaining these weights, the von Neumann-Morgenstern standard gamble, is the posing of a bet where the patient must be indifferent between a certainty of the best value for the measure in question and the worst for all others or a k% chance of the best of all measures and a 100 – k% chance of the worst of all measures, where k is the smallest percentage deemed comfortable by the patient (Figure 1) [28]. The larger this k value, the more preferred the measure. The use of a probability wheel (Figure 2) may aid in minimizing cognitive biases [1, 2] that accompany the elicitation of percentages [5].

Figure 1.

An example of how to assess patient preferences for outcome measures, in this case for “Measure 1.” This is an application of the von Neumann-Morgenstern standard gamble. The patient is given a “Situation A”, with a certainty of the best value for Measure 1 and worst value for all other measures. The patient then assigns the minimum probability, p, for which she would be indifferent between Situation A and Situation B: a p chance of obtaining the best values of all measures and a 1 − p chance of obtaining the worst values of all measures.

Figure 2.

A probability wheel which may be used to minimize the influence of cognitive biases in assessing probability. The assessor is blinded to the numerical value of the probability, instead relying on the area of the pie slice and the random spinner to represent uncertainty.

Risk preference describes the attitude of a patient towards risk. Risk preference can be modeled as a u-curve. This u-curve serves to map any measure’s e-value to a u-value and vice versa. This risk attitude can be modeled as a convex curve, such as an exponential or logarithmic function [29, 30]. If the u-curve is a perfectly straight line, or linear, then

$$u_{\text{linear}}(x)=\frac{x-e_{\text{worst}}}{e_{\text{best}}-e_{\text{worst}}}$$

where e_worst is the worst possible e-value and e_best is the best and we say that the patient is risk neutral with respect to that measure. In such a case, a patient would place a value of $1000 on a 50–50 chance of getting a reconstruction that cost $2000 or a reconstruction that was free ($0).

V. The Breast Reconstruction Decision: A Fictional Case Study

Jenn is a 49-year-old, non-Hispanic Caucasian female with no prior predisposition to breast cancer. She is diagnosed with ductal carcinoma in situ of the left breast. She is otherwise completely healthy, of average weight, with no relevant comorbidities, and a non-smoker. She will receive chemotherapy but not radiation therapy. Let us assume that breast conservative therapy (BCT) is not an option because Jenn adamantly believes that mastectomy eliminates her risk for recurrence whereas with BCT she retains some risk. Even before discussing treatment options for her cancer such as radiation therapy or chemotherapy, she is sent to consult with a plastic surgeon regarding possible breast reconstruction.

Jenn is faced with a breast reconstruction decision. The surgeon informs her that she may elect to have the breast reconstruction performed immediately after the mastectomy or delayed some time after. She may consider a prophylactic mastectomy and reconstruction for the opposite breast, which may be performed at the same time as her afflicted breast or delayed. However, as Jenn’s cancer is sporadic and risk for contralateral disease is small, she may not wish to pursue prophylactic contralateral mastectomy. The plastic surgeon then explains the procedures for the various transverse rectus abdominis muscle (TRAM) flap variants, latissimus dorsi flap variants, gluteal flap variants, tissue-expander and implant, and other less often performed methods. The plastic surgeon then attempts to communicate the outcomes of the various methods and concomitant pros and cons of each method. This consultation takes over an hour of their time [31] and Jenn may leave with no more clarity of action than when she entered. Let us help Jenn make a good decision.

Previously, we have defined, in-depth, the alternatives and possibilities [32]. The breast reconstruction decision problem is depicted as a decision tree composed of nodes connected by branches (Figure 3 and 4). Square nodes indicate decisions over which Jenn has control. Circular nodes indicate uncertain possibilities. The possibility tree is connected to each possible combination of decisions. Jenn has to make three kinds of decisions: (1) laterality, (2) timing, and (3) method (Table 1).

Figure 3.

The breast reconstruction decision tree. Note that the possibilities are not shown. A patient may choose between unilateral, bilateral, or no reconstruction. Then she must decide on the reconstruction timing, whether immediate or delayed, and finally the method of reconstruction. If the patient elects bilateral reconstruction, she must make timing and method decisions for each breast.

Figure 4.

The possibility tree given decisions made concerning the reconstruction laterality, if any, method(s), and timing(s). This tree is connected to the tips of every decision combination in Figure 3. Note that there are 27 different outcomes of breast reconstruction ranging from an excellent/good aesthetic outcome with no complications and no more than one revision to a poor aesthetic outcome with a major complication and five or more revisions.

Table 1.

The kinds and degrees of decision distinctions and their clarifications for breast reconstruction decision-making.

DECISION DISTINCTION		CLARIFICATION
Kind	Degree
Reconstruction Laterality	Unilateral	Only one breast is reconstructed after unilateral mastectomy.
	Bilateral	Both breasts are reconstructed after bilateral mastectomy.
	Neither	Breast reconstruction is not performed after mastectomy.
Reconstruction Timing	Immediate	After mastectomy, while under the same anesthesia, one breast is reconstructed.
	Delayed	After mastectomy, one breast is reconstructed at a later time (not immediate).
Reconstruction Method	Tissue Expander & Implant	One breast is reconstructed using a breast implant to supply the missing breast volume. The breast is reconstructed by the gradual process of hydraulically expanding or stretching the pectoralis major muscle and overlying breast skin so that a breast implant can eventually be placed in the space created by the tissue expander.
	Latissimus Dorsi	One breast is reconstructed using the latissimus dorsi muscle, overlying skin, and a breast implant which may require tissue expansion. Or the muscle (and additional fat tissue to supply additional breast volume) may be used without implants.
	TRAM	One breast is reconstructed using a variant of the transverse rectus abdominis myocutaneous flap, composed of skin, fat, and/or a part or all of the underlying rectus abdominis muscle.

Next, we enter the realm of possibilities. We will consider three kinds of possibilities: (1) number of expected revisions, (2) worst expected complication, and (3) aesthetic outcome. Each of these kinds of possibilities have degrees associated with them. For revisions, we consider three degrees: (1) 0–1 revisions, (2) 2–4 revisions, (3) 5 or more revisions. In complications, we conceptualize them as: (1) none, (2) minor, and (3) major. And for aesthetic outcome, we specify: (1) good/excellent, (2) fair, and (3) poor. These, of course, should also be thoughtfully defined (Table 2). While the distinctions of possibilities and their degrees do not differ between methods, their likelihoods may differ and it is these differences that characterize the alternatives. These possibilities may also provide clues as to how the measures are influenced.

Table 2.

The kinds and degrees of chance distinctions and their clarifications for breast reconstruction decision-making.

CHANCE DISTINCTION		CLARIFICATION
Kind	Degree
Complications	None	No complications.
	Minor	Requires local wound care.
	Major	Requires an invasive procedure or that threatens the patient’s life (e.g., large amounts of mastectomy skin necrosis, hematoma, seroma, rupture/leakage, implant migration, extrusion, displacement, peri-implant calcification, breast asymmetry, infection around implant, toxic shock syndrome, pneumothorax, deep venous thrombosis, pulmonary embolus, pneumonia, respiratory arrest, prolonged ventilation, massive blood loss and need for transfusion, severe sepsis, stroke, myocardial infarct).
Revisions	0–1	None or one procedure to improve the reconstructive aesthetic outcome or correct complications
	2–4	Two, three, or four procedures to improve the reconstructive aesthetic outcome or correct complications
	≥ 5	Five or more procedures to improve the reconstructive aesthetic outcome or correct complications
Aesthetic Outcome	Excellent/Good	7, 8, 9 or 10 on a scale from 0 to 10 (A score of 10 would be considered the perfect reconstruction.)
	Fair	4, 5, or 6 on a scale from 0 to 10
	Poor	0, 1, 2, or 3 on a scale from 0 to 10 (A score of 0 would be considered the worst reconstruction imaginable.)

For every possible outcome, we assign values to each measure. We propose the following seven: (1) out-of-pocket expenses, (2) time lost to surgery, (3) satisfaction with breasts, (4) psychological well-being, (5) chest well-being, (6) abdominal well-being, and (7) sexual well-being. The last five measures can be measured by the BREAST-Q and are further defined in Table 3 [33, 34]. However, fewer measures can be used if Jenn deems certain measures to be unimportant.

Table 3.

BREAST-Q outcome measures and their definitions.

BREAST-Q Outcome Measure	Definition
Satisfaction with Breasts	Describes your overall satisfaction with your current breasts based on how they look, clothed and unclothed, and how your clothing and bras fit. A score of 100 indicates that you are very satisfied with how your breasts look in the mirror with clothes on and off and that you are very satisfied with the comfort of your bras and clothing. If you are less than very satisfied, then your score will be lower.
Psychosocial Well-Being	Describes how you feel, mentally, about your breasts. A score of 100 indicates that, with your breasts in mind, you feel confident about yourself in public, feminine, normal, like other women, and attractive all of the time. If you do not always feel this way about yourself, your score will be lower.
Physical Well-Being: Chest	Describes how you feel, physically, about your breasts. This captures things like real pain or other physical disabilities in and around your breasts such as in the neck, upper back, shoulder, ribs, and arms. A score of 100 indicates that you never have any pain, tightness, or tenderness in or around your breasts. Nor do you have difficulty moving your arms. If you do not always feel pain-free, you would have a lower score.
Physical Well-Being: Abdomen	Describes how you feel, physically, about your abdomen. This captures things like real pain or disabilities in your abdomen and lower back. A score of 100 indicates that you never have difficulty sitting up, doing everyday activities that involve your abdomen like bending over, abdominal discomfort or bloating, or lower back pain. If you have any difficulty sitting up, doing everyday activities because of abdominal weakness, abdominal discomfort or bloating, or lower back pain, your score will be lower.
Sexual Well-Being	Describes how you feel sexually. A score of 100 indicates that you always feel sexually attractive clothed and unclothed, sexually confident, sexually confident when your breasts are unclothed, comfortable during sexual activity and satisfied with your sex life. If you do not always feel this way, then your score will be lower.

At the tips of the tree, we assign e-values for each of the measures for each possible outcome. For each possible branch, we also assign a probability describing its likelihood. The values in the example tree (Figure 5) were randomly generated and should never be used on an actual patient. Real e-values and probabilities may be derived from the literature, data, or surgeon estimates but should be tailored to the particular patient [26, 35].

Figure 5.

How to assign u-values to breast reconstruction possibilities assuming that Jenn is risk neutral and that the additive multiattribute utility function is consistent with her preferences. We only show the process for the first case given a unilateral, immediate, implant/expander breast reconstruction with 0 or 1 revisions, no complications, and an excellent/good aesthetic outcome, starting with the e-values. We also show how to fold-back the tree through probability calculus. Probabilities are shown in italics. Note that we truncate the tree to save space.

The next step is to convert the e-values to u-values. We assess Jenn’s measure preferences using the standard gamble (Figure 1). Let us assume that Jenn is risk-neutral and her preferences for measures are reflected in Table 4. After calculating the u-values for each measure, the next step is to combine these individual u-values into an overall u-value for each possibility. This can be accomplished with multiattribute utility theory [14, 29, 30]. There are a number of candidate multiattribute utility functions (MAUFs) that may be used. We have shown that they perform equally well [36]. Let us assume that the additive MAUF is consistent with Jenn’s preferences. We take the sum of the product of weights of each measure, k, and corresponding u-value, written as

Table 4.

Jenn’s preferences for breast reconstruction outcome measures as weights.

i	MEASURE	WEIGHT (ki)
1	Cost ($)	0.06
2	Time (days)	0.06
3	Satisfaction with Breasts	0.13
4	Psychological Well-Being	0.13
5	Chest Well-Being	0.25
6	Abdominal Well-Being	0.25
7	Sexual Well-Being	0.13

$$\begin{array}{l}u-{\text{value}}_{\text{overall}}=k_{\text{cost}}\times u-{\text{value}}_{\text{cost}}+k_{\text{time}}\times u-{\text{value}}_{\text{time}}+\\ k_{\text{satisfaction}}\times u-{\text{value}}_{\text{satisfaction}}+\\ k_{\text{psych}}\times u-{\text{value}}_{\text{psych}}+\\ k_{\text{chest}}\times u-{\text{value}}_{\text{chest}}+\\ k_{\text{abdominal}}\times u-{\text{value}}_{\text{abdominal}}+\\ k_{\text{sexual}}\times u-{\text{value}}_{\text{sexual}}\end{array}$$

to find the overall u-value for each possibility. However, we must still find the u-value of each alternative through folding-back the tree. We have prepared an interactive spreadsheet example of this process for one reconstruction method (See supplemental digital content 1, spreadsheet, an interactive breast reconstruction decision analysis spreadsheet example for two alternatives: no reconstruction or an unnamed reconstruction method. Please note that the initial values for probabilities and e-values were randomly generated. The reader is encouraged to think of a particular reconstruction method and assign realistic values.).

Probability calculus is used to fold-back the tree resulting in a u-value for each alternative (Figure 5). At the square decision nodes, we pick the alternative with the higher u-value. As Jenn is not having a contralateral prophylactic mastectomy, she has a reduced set of alternatives and we may remove, or prune, branches of the tree that involve bilateral reconstruction. For instance, given a choice between immediate unilateral implant/expander (u-value = 0.60), TRAM (u-value = 0.75), and latissimus dorsi (u-value = 0.70), we pick TRAM (Figure 6). She may choose immediate reconstruction with implant/expander, TRAM, or latissimus dorsi, a delayed reconstruction of the same methods, or no reconstruction at all. The decisions to be made become clear as illustrated by the arrowheads. Using this approach, Jenn should choose a unilateral, immediate, TRAM flap reconstruction, which possesses the highest u-value (0.75) among all the alternatives.

Figure 6.

Jenn’s breast reconstruction decisions, assuming she will not consider bilateral breast reconstruction. Working from right to left, at every square decision node, we pick the alternative for which the u-value is highest as depicted by the arrowheads.

VI. Sensitivity Analysis

Before acting on the best alternative, it may be prudent to individually reexamine some, if not all, of the variables in the decision, particularly those most in doubt. For instance, what if Jenn really has a 0.70 chance of having no complications instead of 0.40? Or what if the possible outcome actually costs $2500 instead of $4500? If the recommended optimal decision changes, then additional sensitivity analysis is warranted for the new best alternative. This requires computational assistance because performing sensitivity analysis by hand would be temporally prohibitive. If the decision remains the same, it is robust and likely good.

VII. Evaluating Decisions: Decision-Quality

A decision cannot be judged by its outcome but by its decision-quality [15, 37–39]. A high quality decision answers the right question (proper frame), has a selection of alternatives that respond to the frame, has reliable information, and considers preferences for possibilities. There must be logic and the patient must be committed towards making the decision.

A validated instrument, the Decision-Quality Scale, quantitatively measures the decision-quality. It consists of ten Likert-like items covering six elements of decision-quality: (1) appropriate frame, (2) creative alternatives, (3) reliable information including models, (4) clear preferences, (5) correct logic, and (6) commitment to action [15, 39]. Examples of items include, “I am having difficulty making decisions about treatment,” “I have a thorough understanding of the medical diagnosis,” and “My doctor and I agree on a treatment strategy [37, 38].”

A qualitative way of illustrating decision-quality is with the decision-quality spider web (Figure 7) [38, 39]. The distance from the inner circle to the outer circle represents the quality of each element. The further from the inner circle, the better the quality. If all the distances are the same and area is maximally large, then the basis is balanced and the decision is good.

Figure 7.

The decision-quality spider web adapted from [39]. Decision-quality is represented by the shaded hexagon, the greater the area, the higher the decision-quality.

VIII. Discussion

We have presented a normative approach to making shared decisions regarding breast reconstruction with an emphasis on patient values. This decision analysis methodology draws upon plastic surgeon expertise in identifying alternatives, literature and data for probability information, and patient preferences for assigning values to outcomes. Other studies in breast reconstruction decision-making have failed to involve actual patients when performing their analyses [40]. Furthermore, we demonstrated the steps necessary to evaluate alternatives, beginning with the e-values of measures, converting them to u-values, using multiattribute utility theory to calculate the u-value of an outcome, and folding-back the decision tree to identify the best decisions. We also discussed how to improve the quality of the decision through sensitivity analysis and balancing the decision basis.

Portions of decision analysis are computationally intensive. A principal obstacle in routine patient decision analysis is the procurement of probability information and e-values for a given patient type and alternative. The large number of these variables underscores the need for automated acquisition from data. These outcome measures do not exist in electronic health records. For instance, measures encompassing quality of life and satisfaction are not routinely obtained from past patients and expenses incurred by an institution [41, 42] are not the same as those borne by the patient.

VIII. Conclusion and Changes in Practice

Decision analysis provides structure for methodic, thoughtful decision-making through the sound use of reason, logic, and mathematics. The goal of decision analysis is not the selection of a good outcome, but of a good choice. Ideally, its application will attenuate worry, safeguard against regret, transmute uncertainty into certainty, and grant some measure of peace in what may ultimately be a very difficult decision.

We believe that decision analysis will lead to (1) a reduction in the average consultation time, (2) fuller involvement of the patient in the decision-making, (3) reduction in additional revisions, and (4) improvement in patient outcomes on average. Once the best alternatives are calculated via decision analysis using the patient’s preferences, the surgeon can focus on educating the patient about a much-reduced set of alternatives. Hence, less consultation time should be required. As we seek to quantify and incorporate the preferences of the patient into the decision-making, we ensure improved participation of the patient. In addition, we may clearly organize what the patient can do, allowing her to focus on the decision aspect rather than the probability aspect, which is not under her control. As a corollary of good decisions, we would expect the number of additional revisions to be reduced on average, possibly due to more realistic patient expectations. As we are using normative decision-making, we would also expect improved outcomes for patients on average by seeking to optimize their psychosocial well-being, financial expenditure, and time commitment. These would be happier patients who should have fewer worries undergoing reconstruction and regrets thereafter.