Explicit criteria for prioritization of cataract surgery

Cataract extraction is the most frequently performed surgical intervention in developed countries [1]. Because of the high prevalence of cataract development and the increasing number of patients requiring cataract extraction, managers must decide how to prioritize patients. Different priority criteria for cataract extraction have been created in the past in different countries, but none has included the appropriateness of the intervention [2–5].

The RAND-UCLA group in the United States [6] developed methodology to establish explicit criteria in the 1980s. This is a very popular methodology that has been used for many different diagnostic and therapeutic procedures since that time and lately for different applications [7–9].

The purpose of the current study was to develop priority criteria, using the RAND method, for patients undergoing cataract extraction and to consider the appropriateness of the indication.

A panel of 11 experts scored 310 scenarios. Figure 2 shows the panel scores for the priority score as a continuous variable. The histogram resembles a normal plot and fulfills normality (p = 0.12 by Kolmogorov-Smirnov test). The panelists' scores tended to be similar (Table 1), except for panelists 5, and 6, who had the extreme high and low mean and median scores.

The logistic regression model indicated that the same three variables were the most important but in a different order, i.e., the visual function, the visual acuity in the cataractous eye, and the anticipated postoperative visual acuity. The other variables were not statistically significant in this model. Appropriateness of the intervention, despite being forced to enter in the model after the rest of the variables to avoid collinearity, remained highly significant and fourth in the order of importance in both models.

We studied the correlation of the scores obtained between the two scoring methods with the priority scores of the original panel. The Pearson correlation coefficient was 0.95 for the optimal scaling model and 0.97 for the general linear model. Using the Spearman correlation coefficient, the results were 0.94 and 0.97, respectively.

Finally, the criteria developed by the panel of experts were summarized by CART analysis on a decision regression tree (Figure 3), which shows that the highest priority was given to the scenarios considered first as appropriate. This is followed by visual function (difficulties with daily activities), visual acuity in the cataractous eye (= 0.4), visual acuity in the contralateral eye, and the anticipated postoperative visual acuity.

We collected information on 936 patients to identify which of the theoretical indications were used in clinical practice. Globally, 102 of 310 theoretical (33%) indications were assigned to 936 patients. By diagnostic group, for simple cataract, 53 of 86 theoretical (62%) indications were assigned to 761 patients with that diagnosis. Of those with diabetic retinopathy, 11 of 118 theoretical (9%) indications were assigned to 19 patients. Of those with other ocular pathologies, 38 of 106 theoretical (35.8%) indications were assigned to 156 patients with that diagnosis.

Cataract extraction has been chosen by different researchers to develop priority scoring systems of [2, 5, 16]. The drawback of all systems is that they do not take into account the appropriateness of the intervention as a previous consideration. Our study incorporates this variable on the priority algorithm, and we show that it plays an important role in patient prioritization.

To develop our system of prioritization, we used the RAND methodology. Although the RAND methodology was developed primarily to create appropriateness explicit criteria, it also can be used for other purposes (9;17;18). To the best of our knowledge, this is the first time the RAND methodology has been used to develop priority criteria linked with previous appropriateness criteria.

We first determined the appropriateness of the indication criteria in two panel rounds. We then selected those indications ultimately classified as appropriate or uncertain. The fact that the appropriate indications were entered in the prioritization algorithm seems logical; however, why should uncertain surgical indications be included? First, because as in some studies in which appropriateness had been related to outcomes, the outcomes of the appropriate and uncertain cases seemed similar [19, 20]. Uncertainty does not indicate that the intervention should not be performed, but rather that either that the panel did not reach agreement or that the panel's opinion is that there is uncertainty about whether the intervention is worthwhile because of lack of evidence or panel experience or ambiguous results. Second, we considered that social dependence, a new variable in our algorithm, might result in an uncertain scenario being considered as appropriate. As other authors who had worked with this methodology pointed out, other circumstances not included in the appropriateness criteria may play an important role in clinician decision making for indicating an intervention. More RAND appropriateness criteria had in common that they include clinically relevant variables but not socially relevant variables or those important to the patient [21]. We considered that social dependence is important in many clinical situations but mainly to prioritize patients. Focus groups performed by our group and other groups studying cataract have shown that patients believe that social dependence is much more relevant than other clinical variables [22]. But also because social dependence has been included in previous priority scoring systems [2, 5].

We initially spent considerable effort defining what we consider priority. Priority differs depending on the health problem and its probability of leading to death or important permanent handicaps if untreated. That introduces the concept of urgency, i.e., if not treated promptly death will result [23], as with myocardial infarction or other cardiovascular problems, in which the risk of death or incapacity is often present. However, cataract itself never implies the risk of death for the patient. In this case, we then defined priority in terms of the benefit of the intervention to the patient. The higher the benefit for the patient, with similar low risks, the higher the priority of the intervention. We understand the benefit as the highest gain in the quality of life or living or working independently for the patient, which would be correlated with relevant improvements postoperatively in their ability to function. With cataract, this improvement should not only correlate with improved visual acuity (either measured by Snellen or lines of improvement or changes in the log MAR visual acuity), as reported by the ophthalmologist, but also with a relevant improvement in visual function, as reported by the patient. If a patient with a preoperative visual acuity of 0.3 and another with 0.1 or less are expected to have improvements in visual acuity to 1.0 and 0.7, respectively, with the rest of their circumstances equal, who should undergo surgery first? The absolute gain in the first case (0.7) is superior to the second (0.6); however, considering functionality, while the visual acuity of the first patient probably imposed limitations, he or she likely could perform some daily activities but was fully functional after surgery. The second patient, who would be legally blind in our country, has a visual function much poorer than the first patient; therefore, a visual acuity improvement to 0.7 may allow him or her to manage daily activities or live independently. The improvement in the second patient has greater relevance considering the quality of life.

As we pointed out previously, most priority criteria developed thus far did not consider the appropriateness of the intervention. Some may assume that all patients on the waiting list are appropriate candidates, which it is not necessarily true. Our results from both the general linear model and the CART analysis indicated that appropriateness was an important variable for the panelists when scoring the priority of each scenario. As reported previously, the rate of inappropriate interventions ranges from 1.3% [24] to 7.7% [25] depending on the study. Although compared to other appropriateness RAND studies those rates are relatively low [24–26], an inappropriate intervention should not be considered for prioritization. However, in the previous studies, uncertain indications ranged from 7% to 36.8%. Because they occur much more frequently, they should be considered carefully.

To the best of our knowledge, one recent study attempted to match appropriateness and priority [3]. However, although the development of appropriateness criteria followed RAND methodology similar to ours, the development of the priority criteria differed and resembled previous studies. Other studies in Canada [2, 16] and New Zealand [27] developed priority criteria by different methods, some of which have been questioned (4). The Canadian group used similar statistical methodology (optimal scaling) to generate a simpler scoring system [2, 16]. In any case, all of them share similar variables in their priority algorithm as ours.

We presented the results of the panel scoring as a categorical variable or continuous variable. To construct the categorical variable, with three categories of priority (from high to low), we used the same algorithm to develop the three RAND categories of appropriateness (appropriate, uncertain, inappropriate). An advantage is that this system may match current priority setting norms in some places such as ours. A disadvantage is that it can seem artificial, or with no conceptual base, to have those categories. On the other hand, the continuous variable provides a priority order for all scenarios, but it can seem less practical. Daily use will determine which one is more useful.

On the other hand, to have an easily usable algorithm, we presented our results simply. We presented two alternative priority scoring systems, one developed by optimal scaling and the other by a classical general linear model. The importance of the variables and the weights assigned to the different categories of variables were similar. In addition, both had very high Pearson and Spearman correlations with the original panel scores. An alternative way to use the real panel scores can be provided by simple software that allows the final score to be obtained by clicking on the categories of each variable. We are working on this alternative method.

Finally, we presented a decision tree that provides a global idea of which clinical situations should be treated first and indicates the importance of assessing the appropriateness of the intervention for prioritization.

In conclusion, this is the first study to combine the appropriateness of an indication with priority criteria following the same methodology. Our results show that appropriateness plays an important role in priority decision making process. The variables included also were identified by other authors, which adds validity to our criteria. We also provide alternative scoring algorithms that allow ease of use of our criteria in clinical practice. Use of them will let us know which one is more useful. Nevertheless, more sophisticated future proposals should include the algorithm in flow models [28] that allows management of patients placed on a waiting list.