What is the best treatment strategy for incidentally detected small renal masses? A decision analysis


Robert Abouassaly, MD, MSc, FRCSC, Assistant Professor, University Hospitals Case Medical Center, 11100 Euclid Avenue, LKS 5046, Cleveland, Ohio, USA, 44106. e-mail: robert.abouassaly@uhhospitals.org


Study Type – Therapy (decision analysis)

Level of Evidence 1b


• To determine the optimal treatment for incidentally detected small renal masses between radical nephrectomy, partial nephrectomy, ablative therapy (AT) and active surveillance (AS) using a decision-analytic Markov model.


• The reference case was an otherwise healthy 60-year-old patient.

• Health utilities and probabilities for postoperative complications, progression to chronic renal insufficiency (CRI), local and systemic recurrence, disease-specific and all-cause mortality were derived from published literature.

• Outcome measures included life expectancy and quality-adjusted life expectancy.

• Extensive sensitivity analyses were performed, including probabilistic sensitivity analyses.


• The mean life expectancy was 18.49 years for partial nephrectomy, 18.09 years for laparoscopic radical nephrectomy, 17.85 years for AT and 17.70 years for AS.

• External validation of our model yielded similar cancer-specific survival rates to the published literature.

• AS became preferred if age at presentation was >74 years, the probability of systemic recurrence on AS was <1.3%/year or when the hazard ratio of death with CRI was >1.63.

• AT became preferred when the probability of systemic recurrence on AT was <1.2%/year, whereas laparoscopic radical nephrectomy was preferred when the risk of CRI with this treatment was <6.6%/year.


• Based on current literature, our model emphasizes the importance of balance between disease control and preserving renal function on life expectancy, and justifies initial active intervention with partial nephrectomy in younger patients.

• Our results are consistent with recent American Urological Association guidelines for the management of this disease.

• However, data used in the model were mostly derived from retrospective data, and thus are subject to selection bias, particularly with respect to AS and AT.


small renal mass


(open) (laparoscopic) radical nephrectomy


partial nephrectomy


chronic renal insufficiency


ablative therapy


active surveillance


life expectancy


quality-adjusted life expectancy


tyrosine kinase inhibitor


hazard ratio.


Although stage for stage RCC is the most lethal genitourinary malignancy, more than 70% of individuals present with localized disease, and those with small renal masses (SRMs) (i.e. tumours <4 cm, clinical stage T1a, American Joint Committee on Cancer 2002 [1]) have an excellent prognosis [2]. In the last three decades, there has been a steady increase in the incidence of RCC [2], predominantly in localized disease [3]. This has been mostly attributed to the incidental detection of SRMs with the increased use of non-invasive abdominal imaging. RCC is most commonly diagnosed in the sixth and seventh decades of life, and notably the greatest increase in incidence has been observed in the later years of life [4].

The standard of care for clinically localized RCC remains surgical resection, resulting in excellent long-term cancer-specific survival. Traditionally SRMs have been treated by open radical nephrectomy (ORN) or, increasingly, partial nephrectomy (PN) [5]. In recent years, ORN is rapidly becoming replaced by laparoscopic radical nephrectomy (LRN) [6], which is associated with less short-term morbidity than ORN but similar impact on long-term renal function [7]. As a result, LRN or ORN may lead to decreased overall survival compared with PN due to greater risk of chronic renal insufficiency (CRI) [8,9]. Ablative therapies (ATs) in the form of cryotherapy and radiofrequency ablation have also emerged as potential treatment options for SRMs [10]. Although early results assessing AT are promising, data on long-term effectiveness and head-to-head studies with surgery are lacking.

Finally, some have argued that incidentally discovered SRMs may not impact negatively on survival, particularly in elderly patients, and as a result active surveillance (AS) has been proposed as a treatment option [11]. Data are emerging, mostly from retrospective case series, on the safety of this approach in selected patients [12,13].

In the present study, we evaluated the optimal treatment strategy for SRMs using a decision-analytic approach. The lack of randomized trials comparing the different treatment modalities in RCC, and the high likelihood that such trials will never be completed, make this question particularly suited for a decision analysis. Using available published data, and taking into account the uncertainty associated with outcomes, a decision analysis allows us to compare approaches and helps determine the optimal treatment modality for patients with SRMs in the absence of comparative randomized controlled trials.


Given that a detailed description of decision analysis and Markov modelling is not feasible within the limits of this paper, we refer readers to a recent review of the use of these methods in urology [14].


Four treatment strategies in the management of SRMs were compared in our analysis: LRN, open PN, AT and AS (Fig. 1A). AT included both cryotherapy and radiofrequency ablation.

Figure 1.

Schematic illustration of decision node (A) and Markov state transition diagrams for LRN (B), PN (C), AT (D) and AS (E).


Our reference case was a 60-year-old patient with an incidentally diagnosed <4 cm solid renal mass suspicious for RCC who would be a candidate for any of the above treatment options. The reason for the size limit is that only renal masses <4 cm (i.e. stage T1a if malignant) are typically considered for PN, AT and AS [15].


Our primary outcome of interest is life expectancy (LE). We chose this outcome since it offers an objective endpoint, is clinically important, and avoids the uncertainties associated with estimated utility values (see below). However, we examined quality-adjusted life expectancy (QALE) as a secondary outcome.


A Markov state transition model (TreeAge Pro Suite 2008, Williamstown, MA, USA) was constructed to compare LE and QALE associated with the treatment of SRMs. The model simulates the natural history of a hypothetical cohort of patients diagnosed with an SRM treated with one of the above-mentioned treatment modalities. After the initial treatment choice, patients can transition between discrete health states. A 1-month cycle length was chosen to reflect short-term treatment-related morbidities [16]. The model was terminated after 480 cycles or age 100 years, after which the degree of uncertainty in most available data renders the output less meaningful. The model incorporated short-term complications and mortality, local/systemic disease recurrence or progression, as well as the development of CRI. An inclusive definition of CRI was used (i.e. an estimated GFR of <60 mL/min) given the evidence that even patients with this degree of CRI are at significantly increased risk of death and cardiovascular morbidity [17].

In the model, patients initially treated with LRN can develop a short-term complication, show no evidence of disease recurrence (well state), progress to systemic recurrence requiring systemic therapy (ChemoTx state), develop CRI, die of RCC, or die of other causes (Fig. 1B). Patients initially treated with either PN or AT can experience any of the health states associated with LRN and can also develop local recurrence. Local recurrences after PN are treated with LRN (Fig. 1C), whereas after AT they are treated with repeat AT in two-thirds of cases and LRN in a third of the cases (Fig. 1D) [18]. Patients initially treated with AS can continue to have stable disease (well state) or experience any of the other health states associated with PN or AT (Fig. 1E). Local disease progression after AS is treated with AT one-third of the time, PN one-third of the time and LRN one-third of the time [19]. Second local recurrences after either AT or AS are treated with LRN. Patients who underwent LRN, PN or AT after initial PN, AT or AS had similar subsequent transitions to those treated initially with the corresponding modality. A major criticism of the AT and AS literature, given that biopsies are not consistently performed, is that their reported outcomes include both benign and malignant renal masses. On the other hand, studies reporting oncological outcomes after RN or PN only include malignant lesions. Pathological data suggest that 20–26.3% of tumours <4 cm are benign [20–22], making comparisons between treatment approaches difficult. Therefore the probability of disease progression in the AT and AS arms was adjusted for the likelihood of having a malignant lesion.


Several simplifying assumptions were made in the construction of the Markov model. First, we assumed that gender did not impact on outcomes after SRM treatment, since there is little evidence that the prognosis of SRMs differs by gender [23]. Second, we assumed that all RNs are done using the laparoscopic approach. LRN is becoming widely accepted as the standard of care for the radical treatment of RCC, particularly when lesions are small [6,24]. We also assumed that all PNs are performed through the open approach. Although laparoscopic PN has recently emerged as a viable option in the treatment of the SRM, there are limited studies with follow-up over 7 years [25]. Moreover, differences in morbidity with laparoscopic PN compared to open PN are relatively minor and probably impact only short-term quality of life outcomes. Third, we assumed that patients who experience local recurrence after either PN or AT could not be salvaged with PN. There is evidence that nephron-sparing surgery (i.e. PN) after either of these treatments is difficult due to scarring and fibrosis [26]. Fourth, we assumed that, after a delay of 12 months (a typical symptom-free period), patients who had developed metastatic disease were all placed on systemic therapy with a tyrosine kinase inhibitor (TKI) until death. Recently, TKIs have become first-line therapy for patients with metastatic disease [27,28]. Although these treatments can result in disease stabilization or partial response, they rarely, if ever, result in a complete disease response or cure and therefore they are frequently continued until disease progression, which is usually a fatal event [27,28]. Fifth, perioperative complications other than CRI only impact the patient for 1 month with no long-term morbidity. Finally, we assumed that CRI after initial treatment occurred at a constant rate over time [7].


A Medline search was conducted for papers published from 1966 to March 2009 to identify probabilities and utilities for use in our model. Combinations of the following medical subject headings and text words or phrases were employed: kidney cancer, nephrectomy, laparoscopy, partial nephrectomy, ablative therapy, active surveillance, complications, renal failure, recurrence, progression, treatment outcome, and survival. Citations were restricted to the English language. We also cross-referenced kidney cancer and surgery with the text word ‘utilities’ to identify utility values associated with RCC and surgical intervention. Textbooks on kidney cancer treatment were consulted and bibliographies of key articles were searched for additional references. Content experts in urological oncology were also consulted to ensure completeness.


The age-adjusted risk of death from other causes was estimated using actuarial life-tables [29]. Per-cycle probability estimates were obtained by converting the published probabilities to rates, and then converting these to monthly probabilities assuming an exponential distribution [P(x) = 1 − euT  ]. The probabilities of treatment complication and mortality, progression to CRI, and disease recurrence were mostly obtained from large retrospective case series (Table 1[6,7,10,12,15,17,20–22,29–52]). Probability estimates from prospective series were used when available. Since standard deviations were rarely reported in the included studies, when multiple values for probability estimates were obtained, the final probabilities were calculated by weighting each paper’s probability by the inverse of the variance. Plausible ranges for each probability were defined as the extremes of published values where multiple studies were used to obtain a pooled probability estimate. When estimates were obtained from a single source, the plausible range was defined as the published 95% CI. When the 95% CI was not available, the range was defined as ±50% of the estimate. The small size of published series examining AT made the accurate quantification of 30-day mortality impossible; therefore an estimate was obtained from a large series of AT in liver tumours with a wide plausible range (i.e. from zero to the upper estimate for LRN). Furthermore, since no data are available on delay of systemic therapy after detection of progression, an estimate of 12 months was chosen based on consultations with local expert/experienced medical oncologists. Once again, a wide plausible range was used for sensitivity analyses (i.e. 1–36 months).

Table 1.  Model probabilities and utilities, and one-way sensitivity analysis results on applicable variables
  1. All probabilities are given as 1-year estimates where applicable. Preferred strategies below and above threshold value are listed from left to right. *From actuarial life-table.†Upper bound defined as upper 95% confidence limit; lower bound assumed to be no lower than that of surgical interventions.‡Upper bound of range for treatment interventions assumed not to be higher than that of AS; lower bound assumed to be no lower than that of surgical interventions.§PN is preferred unless otherwise specified.

Age at diagnosis, years [6]6030–9073.6 (PN/AS)65.5 (PN/AS)
Non-cancer death [29]Age-specific table*  
  Hazard ratio for death from CRI [17]1.21–5.91.63 (PN/AS)1.35 (PN/AS)
  During chemotherapy phase [30]0.2630.223–0.303
  AT (given short-term complication) [31]0.07240–0.302
  PN (given short-term complication) [32–34]0.0420–0.107
  LRN (given short-term complication) [32]0.09770–0.186
 AT [35]0.01670–0.09950.00527 (AT/PN)
  PN [7]0.07190.0527–0.09950.0987 (PN/AT)
  LRN [7]0.2950.0273–0.3480.0664 (LRN/PN)0.0701 (LRN/PN)
 Local recurrence    
  AS [36]0.02770–0.0730 [12,37]N/A0.0536 (AS)
  AT [15]0.03040.0182–0.0427
  PN [15]0.005860.00484–0.00684
Systemic recurrence    
  AS [36]0.01900.0113–0.03980.0132 (AS/PN)0.0171 (AS/PN)
  AT [15]0.01580.0113–0.03980.0117 (AT/PN)0.0142 (AT/PN)
  PN [15]0.01270.0113–0.0142
  LRN [15]0.01270.0113–0.0142
 Adjustment for unknown histology of SRM [20–22]0.7540.737–0.80
 Short-term complications    
  AT [10,38,39]0.0530–0.088
  PN [33,40–42]0.150.041–0.386 [43]
  LRN [44,45]0.0860.03 [46,47]
 Delay in initiating TKI after systemic recurrence12 months1–36
 AS, assumed10.68 [48]–1N/A
 Chemotherapy0.74 [27]0.42–0.9 [49]N/A
 CRI0.9 [49]0.7–0.9 [49]N/A0.86 (AS/PN)
 Short-term complication0.67 [50]0.18 [51]–0.99 [52]N/A


The preference that society or individuals have for a given health outcome is referred to as utility. A utility of 1.0 typically represents perfect health, while 0 represents the worst health state (e.g. death). For our model, published utility values for RCC could not be located; as a result, utility values were extrapolated from clinically comparable medical conditions. The utility value of AS was assumed to be 1.0; this was the value assigned to patients with prostate cancer on watchful waiting in a previously published decision analysis [50]. The utility of having metastatic RCC treated with a TKI was estimated by standardizing the total Functional Assessment of Cancer Therapy – General score of patients with metastatic RCC on a TKI to a 0–1 scale [27,53]. Utility estimates for other chemotherapeutic regimens were used to create a plausible range [49]. No published utility estimates exist for RCC surgery; therefore an estimate was derived from a decision analysis assessing treatment options in another genitourinary malignancy [50], and once again a wide plausible range was used to account for the uncertainty in the estimate.


Given the tendency to use patient age to determine suitability for intervention in clinical practice [54], we assessed the effect of varying age at presentation on the preferred treatment strategy.


Model-predicted survival outputs were compared with external series as a method of model validation.

One-way sensitivity analyses were performed on all variables in the model within their plausible ranges to determine the effect on the preferred treatment strategy. Two-way sensitivity analyses were conducted on clinically meaningful combinations of variables.

In order to assess the joint uncertainty of all model parameters simultaneously as well as provide a more accurate estimate of the average health outcomes of each strategy, we performed probabilistic sensitivity analysis using 1000 second-order Monte Carlo simulations. Parameter estimates and standard errors were derived for all variables. Transitional probabilities were assumed to follow a gamma distribution (bounded by zero and infinity), whereas event probabilities and utilities were modelled using a beta distribution (bounded by 0 and 1). In each run of the model, we sampled each event probability and utility from the respective parameter distributions following standard convention [55].



The mean LE of a 60-year-old patient with an incidentally diagnosed SRM treated by either PN, LRN, AT or AS was estimated to be 18.49, 18.09, 17.85 and 17.70 years, respectively. PN offered an incremental LE gain of 4.8 months compared with LRN, 7.7 months compared with AT, and 9.5 months compared with AS (Fig. 2). When factoring in utilities, the QALE estimates for PN, LRN, AT and AS were 17.40, 16.40, 17.12 and 17.11 years, respectively. The latter estimates did not change the primary strategy (PN) but yielded a reversal in the strategy ranked second, with incremental gains of 11.9 months for PN compared with LRN and 3.2 months compared with AT or AS.

Figure 2.

Incremental LE and QALE gains for PN, AT and AS compared with LRN.


Survival estimates from our model for the LRN and PN arms were compared with 5- and 10-year cancer-specific survival rates from a study assessing long-term oncological outcomes using these treatments [56,57]. Our model yielded 5- and 10-year estimates of 97.3% and 92.6% for LRN, and 97.2% and 92.4% for PN, respectively. These values compared favourably with published values of 95% and 95% for LRN and 97.6% and 94.5% for PN, respectively.


Our model indicated that although PN was preferred in patients younger than 74 years of age, in older patients AS was preferred, offering a small advantage in LE (Fig. 3a). When QALE was assessed, the age threshold where AS was preferred over PN decreased to 66 years (Fig. 3b).

Figure 3.

Incremental LE (A) and QALE (B) gains by age at presentation for PN, AT and AS compared with LRN.


In terms of LE, one-way sensitivity analyses showed that our model was sensitive to the increased risk of death associated with CRI, the probability of renal failure after LRN, and the probability of systemic progression after AT and AS. AS was preferred when the hazard ratio (HR) of death with CRI was >1.63 or the probability of systemic recurrence was <1.3%/year, AT was preferred when the probability of recurrence was <1.2%/year, whereas LRN was preferred when the risk of CRI with this treatment was <6.6%/year. Sensitive variables as well as threshold values for QALE are given in Table 1.

Two-way sensitivity analysis revealed that, as age at diagnosis increased and the HR of death with CRI increased, AS became favoured over PN. When the HR of death with CRI was set at the extremes of 1.0 and 5.9, AS became the preferred strategy for patients over age 86 years and 35 years, respectively. Only when the HR of death with CRI was low and the risk of CRI after LRN was <6.6%/year did LRN yield the highest LE.


Jointly varying the input parameters in a probabilistic manner did not alter the treatment decision. The mean LE for the reference case by probabilistic sensitivity analysis was 19.0 (95% credible interval 17.3–21.2) years for PN, 18.5 (16.3–21.2) years for LRN, 18.5 (15.3–21.3) years for AT and 18.5 (15.1–21.5) years for AS. The wide credible intervals, particularly with AT and AS, reflect the uncertainty associated with the model parameter estimates. Simulations favoured PN 35.3% of the time, AS 34.7% of the time, AT 20.1% of the time and LRN 9.2% of the time.


To our knowledge, ours is the first decision analysis attempting to determine the optimal management of an incidentally detected SRM. Using the best available published data, our model determined that, for a 60-year-old patient with an incidentally diagnosed SRM, the LE was approximately 5 months higher for patients treated with PN compared with LRN, and even higher compared with AT or AS. On the other hand, in patients older than 74 years of age the favoured treatment strategy became AS. This is consistent with clinical practice, where AS is generally reserved for patients with advanced age (usually >70–75 years) or comorbidity with a limited life expectancy and elevated surgical risk [11,58].

There is increasing evidence that CRI is associated with increased mortality and cardiovascular morbidity [17,59–61]. It is also clear that LRN is associated with a greater risk of CRI compared with PN [7]. Furthermore, there are suggestions that the decreased renal function associated with RN may compromise survival [8,9]. This is also reflected in our findings, where renal function preservation appears to be the most important determinant of LE. Our model suggests that PN should be preferentially used in young patients. Even when the joint uncertainty in our estimates is taken into account, PN remains the favoured strategy. In fact, recently published clinical guidelines from the AUA for management of SRMs suggest that PN be considered the standard of care, and that RN should be reserved for those in whom PN is not technically feasible [62]. Our model suggests that LRN is only favoured when the rate of progression to CRI is low. Although our findings are not reflected in/by clinical practice, they confirm the suggestion that PN may be under-used in the treatment of RCC [63].

We did not examine ORN as a treatment option, given the decreased morbidity and equivalent 10-year oncological outcomes with LRN [57]. Although we acknowledge that currently many RNs are still being done through a traditional incision, we wanted to examine the best case scenario, and ORN would clearly not offer an advantage in terms of LE or QALE compared to LRN. Similarly, we did not include laparoscopic PN in our model. Although the laparoscopic approach offers less short-term morbidity than open PN, its impact on long-term recurrence and renal function is currently unclear.

A major difficulty in building a decision model for the treatment of SRMs is the paucity of prospective clinical trials. Most studies examining these approaches are retrospective case series, which are subject to selection and reporting biases and probably under-represent patients who do poorly. Although both AT and AS were included in the decision analysis, it is generally felt that estimates of progression in the published literature are preliminary with shorter follow-ups compared with studies examining RN or PN. In addition, studies of AT or AS are subject to strong selection bias. Tumours treated in this manner are typically smaller, and lack aggressive features on imaging or biopsy [15]. Thus, it is likely that our model overestimates the oncological efficacy of AS and AT, and as a result underestimates the true advantage of PN over these modalities. We are aware of only one unpublished prospective study evaluating a cohort of patients on AS, and currently length of follow-up is limited (i.e. 13 months) [36]. Our results are highly sensitive to changes in the probability of either local or systemic disease progression on AT or AS. As such, additional high-quality outcomes data for these treatment modalities will be crucial to informing future decision making.

Despite the fact that disease recurrence after treatment of SRMs is rare, our model attempts to simulate how these patients are managed in clinical practice. This was particularly difficult in the case of AT and AS where published data on re-treatment are limited. Given the evidence that PN after AT is challenging [26] and the fact that most local failures of AT are re-treated with repeat AT [18], we modelled local re-treatment as two-thirds undergoing repeat AT and one-third LRN. Similarly, using the best available evidence and taking into account the fact that local progression while on AS would limit nephron preservation in some [19], we thought it reasonable to model treatment after AS as one-third undergoing AT, one-third PN and one-third LRN. Furthermore, in an attempt to better simulate clinical practice, TKI therapy was not started at the first sign of systemic spread; rather, treatment was delayed for 12 months on average. This is because TKI therapy has significant toxicity and is often not immediately initiated in asymptomatic patients with minimal systemic disease. However, since no published data exist on the average treatment delay, we performed a sensitivity analysis with a wide plausible range (i.e. 1–36 months). Not surprisingly, since systemic recurrence after treatment of SRMs is a rare event, this did not alter the preferred treatment strategy in our model.

There are several strengths to our model. This is the first model-based analysis of the optimal treatment of SRMs. We included the main treatment strategies, including AT and AS. We also carefully modelled the long-term impact of CRI using current data. Finally, our model produced survival estimates similar to those in the published literature, supporting its validity. One of the advantages of decision analysis is that it highlights areas of uncertainty quantitatively and points out the need for high-quality outcomes data in the treatment of SRMs, particularly for AT and AS. However, there are important limitations to our study that need to be acknowledged. First, perhaps the greatest limitation is the lack of high-quality outcomes data from randomized trials in RCC. In addition, published data are often from tertiary and quaternary referral centres and may not reflect outcomes in general clinical practice. As a result, we used relatively wide plausible ranges to account for the uncertainty associated with our probability estimates along with performing extensive sensitivity analyses. As better long-term data become available, particularly for AS and AT, our probabilities will need to be modified to better reflect reality. Second, another important limitation of our model is that patient comorbidities have not been considered, primarily because of the absence of outcomes data for several treatments stratified by comorbidity. Comorbidity would clearly have an effect on treatment choice, particularly in older patients, by impacting death from other causes as well as treatment-related morbidity. Finally, there is a paucity of validated utility estimates in RCC. Once again, this limitation was overcome by using wide plausible ranges in our sensitivity analyses to account for this uncertainty. Additionally, we focused on LE as our primary outcome, but included QALE outcomes to allow readers to judge for themselves the value of adding utility information. It is also reassuring that our findings were not substantively changed with the incorporation of utilities.

Our model indicates that in a 60-year-old patient with an incidentally detected SRM, PN offers higher LE than LRN, AT or AS. In older patients, however, consideration should be given to less invasive techniques with the goal of renal function preservation as this appears to offer the best LE and QALE. Our results are consistent with recent AUA guidelines for the management of SRMs [62]. The retrospective nature of the data used in the model may overestimate outcomes, particularly for AS and AT. Thus, the future availability of long-term data on AT and AS may result in a change in the preferred treatment for SRMs.


Robert Abouassaly was awarded a research fellowship from the Canadian Cancer Society. We would also like to acknowledge Ahmed Bayoumi for his instructions on decision-analytic techniques and Markov model building.


None declared.