Nonmalignant mammary tumors (NMT) are common in intact female dogs. Little is known about the clinical significance of these tumors, and the effect of ovariohysterectomy (OHE) on their development.
Nonmalignant mammary tumors (NMT) are common in intact female dogs. Little is known about the clinical significance of these tumors, and the effect of ovariohysterectomy (OHE) on their development.
Ovarian hormone ablation through OHE decreases the risk of new tumors and thereby improves long-term prognosis for dogs with NMT.
Eighty-four sexually intact bitches with NMT.
Dogs were allocated to undergo OHE (n = 42) or not (n = 42) at the time of NMT removal in a randomized clinical trial. Tumor diagnosis was confirmed histologically in all subjects. Information about new tumor development was collected via follow-up phone calls and recheck examinations. Separate survival analyses were performed with the endpoints new tumor development and death. Cause of death was classified as related or unrelated to mammary tumor. In addition to OHE status, the influence of age, body weight, breed, tumor size, tumor number, tumor duration, type of surgery, and tumor histology was investigated.
New mammary tumor(s) developed in 27 of 42 (64%) intact dogs and 15 of 42 (36%) ovariohysterectomized dogs (hazard ratio 0.47, P = .022). Nine of the 42 dogs (21%) which developed new tumors were euthanized because of mammary tumor. Survival was not significantly different between the 2 treatment groups. In the intact group, nine dogs subsequently developed ovarian–uterine diseases.
Ovariohysterectomy performed at the time of mammary tumor excision reduced the risk of new tumors by about 50% among dogs with NMT. Survival was not significantly affected. Adjuvant OHE should be considered in adult dogs with mammary tumors.
benign mammary tumor
nonmalignant mammary tumor
Mammary tumor is the most common tumor in sexually intact female dogs.[1-5] Close to 60% of these tumors are benign. Only 1% of mammary tumors smaller than 1 cm are malignant and 50% of tumors larger than 3 cm are benign. The clinical implications of benign mammary tumors (BMT) are still largely unknown and are commonly considered a disease of limited clinical importance. This is in contrast to malignant mammary tumors with potential for metastasis and fatal outcome.[8-10] However, BMT is a recognized risk factor for developing additionally mammary tumors later.[11, 12] Mammary tumors in dogs develop as a histologic continuum from benign to malignant, with the malignant tumors representing the end stage of this continuum. It therefore seems plausible that a subset of dogs with benign tumors could be at risk of developing malignant, and thus potentially fatal, tumors. In comparison, women with a benign breast tumor or atypical mammary hyperplasia have increased risk of subsequently developing breast cancer, and this risk increases with increasing atypia. Factors leading to the subsequent development of malignant neoplasms have not been studied in dogs with BMT.
One well-recognized risk factor for mammary tumor development is exposure to ovarian hormones during the 1st 2 years of life, as demonstrated by the significantly reduced risk of mammary tumors in dogs that undergo ovariohysterectomy (OHE) during this time period. OHE performed after the age of four has been found to have limited to no protective effect against mammary tumors.[14-16]
New tumors in other mammary glands are common in dogs with a previous diagnosis of mammary tumors.[7, 17] One of 4 dogs develops new tumors after surgical removal of BMTs.[7, 18] And importantly, these later tumors can be malignant even if the initial tumor was benign,[17, 18] and thus might pose a risk of metastasis and premature death for the affected dog. Several studies have shown that a significant portion of the benign epithelial tumors and hyperplastic mammary lesions express estrogen and progesterone receptors, inferring continued hormonal dependence.[19, 20] We therefore hypothesized that ovarian hormone ablation through OHE would decrease the risk of new tumors and thereby improve long-term prognosis for dogs with nonmalignant mammary tumor(s) (NMT). A recent meta-analysis on the protective effect of OHE on the risk for mammary tumor development using Cochrane principles concludes that the level of scientific evidence is too weak to firmly conclude on this issue. Therefore, a well-controlled randomized study to answer the questions regarding the effect of OHE in dogs with mammary tumors is clearly needed.
The purpose of this study was to determine the effect of OHE on new tumor development and survival in dogs with NMT. A secondary objective was to determine the association between the initial and subsequent tumors regarding histology and tumor localization.
The study was designed as a randomized controlled clinical trial. Intact female dogs with mammary tumors and no previous history of mammary malignancy were eligible for inclusion. A complete reproductive and health history was collected as part of the screening and initial enrollment. Randomization to either tumor removal or tumor removal and concomitant OHE was performed before surgery. In addition, the dogs were stratified based on tumor size (<3 cm, ≥3 cm) and age (<9 years, ≥ 9 years) to ensure equal distribution of these 2 prognostic factors between the treatment groups. A block randomization scheme was used within each stratum. The allocation sequence was computer generated, and the treatment allocation was not known to owners or the investigator until the enrollment had been completed.
All owners of participating dogs signed a written consent form where they were given relevant information about the project and agreed to the randomization procedure. There was 1 fixed price for all participants irrespective of type of mammary tumor surgery or if adjunctive OHE was performed or not. The study was approved by the institutional animal care and use board at the Norwegian School of Veterinary Science.
Complete staging, including blood work (complete blood count and serum biochemistry profile), cytological investigation of enlarged draining lymph nodes, and 3-view thoracic radiographs, was performed before surgery. Dogs with distant metastasis or other serious diseases were excluded. All tumors were recorded (number, localization) and size was determined as the largest diameter measured by caliper. Surgery was performed according to standard surgical practices[22, 23] and involved excision of at least 1 cm of gross normal appearing tissue surrounding the tumor dependent on size and number of the mammary tumors treated. No prophylactic mastectomy of normal glands was performed. Age, weight, and breed of the dogs were recorded at the time of surgery. OHE was performed concomitantly with tumor surgery.
Based on results from the histology review, 2 separate groups were created: a carcinoma group consisting of dogs with at least one malignant epithelial tumor, and a nonmalignant group which included dogs with hyperplastic nodular lesions, benign tumors regardless of tissue of origin, and carcinoma in situ. Here, we report on this latter group of dogs. Both tumors and hyperplastic lesions in the nonmalignant group are termed nonmalignant tumors (NMT) because they are clinically indistinguishable from each other when appearing as discrete well defined “tumorlike” lesions. Cases were enrolled continuously until an adequate number of dogs with malignant epithelial tumors were enrolled in the malignant group. The results from this group will be reported separately.
After surgery, the owners of the dogs were instructed to monitor for any signs of new mammary tumors and notify the principal investigator (PI) if any signs of recurrence or new tumors were noted. In addition, they were contacted by the PI (VK) every 6 months through phone to ensure this information regarding recurrence or new tumor development (localization and time). Other health issues were recorded as well. Dogs with reported/suspected new tumors were requested to return for clinical examination and confirmation. Surgical excision was recommended for all dogs with new tumors; however, no financial study support was provided for any of these subsequent treatments. Complete necropsies were requested regardless of cause on dogs that died or were euthanized.
All tumors were fixed in 10% neutral buffered formalin and submitted for histopathological examination. The slides were stained with hematoxylin and eosin. To minimize bias because of misdiagnosis, the slides were evaluated by 2 independent pathologists (MHG and JT) who performed the evaluations blinded to each other's diagnoses. Dogs diagnosed with only NMT by at least one of the pathologists, and subsequent agreements, were included. The tumors were further classified according to type of tissue present (epithelial, myoepithelial and/or connective tissue). A complete histopathological description was provided for each tumor and included information regarding the degree of tumor differentiation, presence of cellular atypia, carcinoma in situ, and tumor margins.
The 2 main outcome variables in the analysis were time to 1st new mammary tumor and time to death/euthanasia. Cause of death/euthanasia was also recorded.
New tumors were defined as any tumor arising in the mammary tissue after excision of the original tumor(s). Tumors occurring in the same gland or close to the original tumor were also classified as new tumors, rather than recurrences. All subsequent tumors detected by owner, during rechecks and/or at necropsies were recorded. If the dogs were not available for recheck examination, information about new tumors was based on owner's description. Localization of new tumors was recorded referring to closest associated mammary gland. Their relation to the original removed tumors were also recorded in terms of adjacency and whether the new tumors were located ipsi- or contralateral to these. The new tumor was, if available for histology, diagnosed by the same pathologists and categorized as benign or malignant.
Cause of death (including euthanasia) was classified as mammary tumor specific or not-mammary tumor specific. To be classified as mammary tumor specific, the owners' decision to euthanize had to be directly related to the tumor, or its metastases, and had to be confirmed by necropsy or by diagnostic imaging. Cause of death in dogs with mammary tumors in which these tumors did not cause clinical problems was classified as not-mammary tumor specific.
In addition to the intervention variable (spayed or not at the time of tumor removal: OHE/non-OHE), the following variables were examined for potential influence on new tumor development and overall survival: age at the time of surgery as continuous and categorized (<9 years, ≥ 9 years) variable, body weight (<22 kg, ≥22 kg), breed (pure, mixed), tumor duration (<6 months, ≥6 months), number of tumors (multiple, single), tumor size (<3 cm, ≥3 cm), extent of surgery (lumpectomy/simple mastectomy, regional mastectomy/radical mastectomy), and tumor histology. Tumor histology was categorized into (1) hyperplasia and cysts; (2) benign tumors (adenomas, complex adenomas, benign mixed) without atypia; and (3) benign tumors with atypia, necrosis or with carcinoma in situ. In dogs with multiple tumors, the tumor with the most atypical histological changes determined the category.
The distribution of potential risk factors by each OHE group was calculated and compared for the variables age, body weight, breed, tumor size, tumor histology, number of tumors, type of surgery, and tumor duration to ensure similar group characteristics at baseline, using chi-squared test. When categorized, the cut-off values were determined by the median derived from all dogs. For age and tumor size, however, the predetermined cut-off values for the stratification procedure were used. Dogs lost to follow-up or still alive without any of the events, were censored at the date of last known status. Dogs that developed ovarian or uterine disease requiring OHE were censored at the date of such surgery. The Kaplan–Meier method was used to compute survival curves and estimate remission–and survival time of dogs by OHE group. Differences in survival between different groups were tested using the log-rank test.
To evaluate the effect of OHE on the two events (1) new tumor development; and (2) death/euthanasia, whereas adjusting for other possible risk factors, separate Cox proportional hazards models were built for each of the events. Two outcomes were modeled for death/euthanasia: death of any cause (overall survival) and death attributable to mammary tumor (tumor-specific survival). Time at risk was defined as months from surgery date to the event or censoring. All the clinical variables were initially screened for effect on tumor development or survival by applying univariable Cox proportional-hazards models with adjustment for age and using a cut-off of P < .10 for offering variables to the multivariable model. Predictors were retained in the final model if P < .05 or if assumed to have great biological interest a priori. Finally, an estimate of the baseline hazard was derived, conditional upon the set of coefficients in the model.
The assumption of proportional hazards was evaluated based on Schoenfeld residuals for the variable OHE in each of the 3 models. If this assumption was violated and graphical assessment indicated a time-varying effect (TVE) of this variable, an interaction term between this variable and time (on either a linear or logarithmic scale) was included in the model. The importance of the assumption of independent censoring was evaluated by sensitivity analyses based on both complete positive and complete negative correlation between censoring and outcome. The amount of explained variation was evaluated by an overall r2 statistic for proportional hazard models. Plots of the deviance residuals, score residuals, and scaled score residuals against time at risk were used to identify outlying observations with influence on the model, and the model was fit with and without any outlying observations. All analyses were performed using the software package Stata.1
Of 330 dogs initially screened for eligibility, 84 dogs had nonmalignant tumors and were included in the study (Fig 1). Histopathologically, 108 (51%) of totally 210 tumors were classified as complex or mixed tumors. Forty-four (21%) tumors were adenomas or variants thereof, and 4 (2%) were diagnosed as carcinoma in situ. The remaining 54 (26%) tumors were diagnosed as different types of hyperplastic or dysplastic lesions. All dogs were evaluated, staged, surgically treated, and whenever feasible rechecked at the Department of Companion Animal Clinical Sciences at the Norwegian School of Veterinary Sciences in the period from February 2005 to May 2012. The median follow-up time for all dogs, including the censored ones was 31.5 months (range 3.5–87.5). For the dogs that were still alive at the end of the study (n = 24), the median follow-up was 34 months (range 19.0–77.9).
Forty-two dogs underwent OHE concomitantly with tumor surgery and 42 remained sexually intact. For the non-OHE group, the median follow-up time was 31.2 months (range 3.5–87.5) and for the OHE group, 31.8 months (range 4.2–78.2). There was no difference between the treatment groups in terms of signalment, tumor characteristics, and type of surgery, see Table 1. Postsurgically, all owners were able to provide follow-up information on their dog by the regular phone-interviews. In addition, 1 or more clinical rechecks, necropsy, or both were performed in 36 dogs in the nonOHE group, and in 34 dogs in the OHE group during the follow-up period.
|Non-OHE Group (n = 42)||OHE Group (n = 42)||P value|
|≥ 9.0 years||19||24|
|Small breed versus other||1.00|
|Small breed (≤10 kg)||7||7|
|Other breed (>10 kg)||35||35|
|Pure versus mix breed||.29|
|Type of surgery||.12|
|Benign with atypiag||7||6|
Dogs from the non-OHE group developed significantly more new tumors (P = .022) than the dogs from the OHE group. Of the 42 dogs that developed new tumors, 27 (64%) and 15 (36%) were from non-OHE and OHE group, respectively.
Twenty-six of the 27 dogs in the non-OHE group with new tumors, and all the 15 dogs in the OHE group had their tumors confirmed clinically by a veterinarian. Only in 1 dog reported to have new tumor, this information was based solely on owners' description. Median time to new tumor development was 20.8 months (range 3.8–80) and 19.6 months (range 2.3–47.2) for the non-OHE and the OHE group, respectively. The Kaplan–Meier survival curves according to OHE status with new tumor as endpoint (Fig 2) illustrate a protective effect of OHE. The OHE variable was found to be statistically significant (P = .022) in the univariable screening with a protective effect of OHE (HR 0.47, 95% CI 0.25–0.89). None of the other variables analyzed were statistically significant in the univariable screening. In the final Cox proportional-hazards model, OHE-status was the only variable that had a statistically significant effect on the hazard of new tumor with a hazard ratio of 0.47 (95% CI 0.25–0.90, P = .022) for the OHE group. However, age at time of surgery was added to the model because of its possible biological importance. The hazard ratio for age was 1.12 per 1-year increase (95% CI 0.94–1.33, P = .27).
The protective effect of OHE against new tumor development was found to vary with time. It increased on a logarithmic time scale (log-time). Thus, an interaction between OHE status and log-time was included in the model. The protective effect of OHE appeared approximately 5 months after surgery and increased up to 48 months after surgery when it reached a more stable level (Table 2).
|Months after OHE||5a||12||24||36||48||60|
By the end of the observation period (May 31, 2012), 60 (30 non-OHE and 30 OHE) of the 84 dogs had died; of these 33 underwent necropsy. Figure 3 illustrates the Kaplan–Meier survival curves by OHE status with death/euthanasia of any cause as the endpoint. There was no statistically significant difference in overall survival between the 2 groups. Median time to death was 31.2 months (range 3.5–87.5) and 31.0 months (range 4.2–78.2) for the non-OHE and the OHE group, respectively. However, the hazard of dying regardless of cause was influenced by the dogs' age at the time of surgery with a hazard ratio of 1.4 per 1 year increase in age (95% CI 1.19–1.62, P < .001). It did not differ significantly between the other investigated variables, including the OHE variable when applying a Cox proportional-hazard model with adjustment for age. The same result was seen when the endpoint was death from mammary tumor.
Nine of the 42 dogs (21%) that developed new mammary tumors were euthanized because of clinical problems related to it (metastatic disease [n = 6], large or ulcerated tumor compromising normal function [n = 2], development of inflammatory carcinoma [n = 1]). Six dogs belonged to the non-OHE group, and three dogs to the OHE group (95% CI: 0.098–1.59, P = .19). In Figure 4, clinical outcome by OHE-status is summarized for new tumor development, death because of mammary tumors, and death related to other causes. Of deaths related to other causes, 19 were because of other tumor types and 32 were related to nonneoplastic causes (multifactorial causes [“old”; n = 10], chronic degenerative joint disease [n = 8], heart disease [n = 4], acute abdomen [n = 3], trauma [n = 3], pyometra [n = 2], kidney failure [n = 1], and idiopathic epilepsy [n = 1]). The assumption of proportional hazards was valid for the OHE variable in the model when death/euthanasia was the endpoint.
In total, 42 dogs (50%) developed new tumors during the follow-up-period. Only 4 dogs had the new tumors surgically excised, but histopathological analysis was available in a total of 27 dogs because an additional 23 dogs were returned for euthanasia and necropsy. Sixteen dogs had at least 1 malignant new tumor and 10 dogs were found to have only benign new tumors. The distribution of these dogs by OHE status is shown in Figure 5.
Of the 12 dogs which initially had 1 single tumor, 7 dogs (58%) developed new tumors ipsilaterally, of which 5 appeared in the adjacent gland. The other 5 dogs (42%) developed new tumors in the contralateral chain, of which 4 were found in the most adjacent of the contralateral glands.
Thirty dogs with new tumor development had multiple tumors initially, thereof 22 cases with bilateral distribution. In the remaining dogs where the initial tumors were unilateral (n = 8), 5 developed new tumors in the contralateral mammary chain.
Six dogs in the intact group developed pyometra for which 3 were treated with OHE (censored at that time) and 3 were euthanized. Moreover, clinically significant uterine/ovarian tumors were found in 2 other dogs; 1 dog developed uterine leiomyosarcoma and was ovariohysterectomized because of this (censored at that time). A 2nd dog was euthanized as a result of acute abdomen because of torsion of a cystic ovary. Also, on necropsy, an ovarian granulosa cell tumor and a vaginal leiomyofibroma were revealed in one of the dogs.
This study shows a significant risk reduction for new mammary tumor development in dogs treated with OHE adjuvant to NMT removal. Based on previous publications,[14-16] it has been assumed that OHE later in life has little effect on mammary tumor risk. However, none of these former studies were prospective, randomized, or stratified according to malignancy nor was the development of new tumors used as endpoint in the analysis. These findings strengthen the similarities between women and dogs in term of the role of hormonal exposure and breast cancer risk; in women the breast cancer risk is associated with the cumulative life time exposure of breast tissue to hormones, our results suggest a similar situation in the dog, the effect of hormones are continuous and contributes to increased risk also after the age of 2.
The hazard ratio of 0.47 (OHE/nonOHE) for new tumor development reflects that bitches spayed at the time of tumor removal, on average run approximately half the risk of developing new tumors at any given time compared with the bitches remaining intact. A hazard can therefore be interpreted as an instantaneous risk, and the term risk will be used in the following discussion because it is probably a more familiar concept to most readers.
The advantage of OHE on time to new tumor development did not become apparent until 5 months after the OHE intervention. This somewhat delayed effect and then increasing benefit over time makes sense from a biological point of view and probably reflect the time needed for ovarian hormones in mammary tissue to be washed out in the dogs undergoing OHE. Also, the fact that most of the dogs in the non-OHE group had time to experience estrus during the 5 months period might explain the increasing benefit of OHE over time. The temporarily increased endogenous hormonal exposure might have accelerated new mammary tumor development in the non-OHE group compared with the OHE group. This may also explain why this benefit becomes apparent after 5 months. Normal mammary epithelial tissues and most benign tumors express hormone receptors. With time, continued exposure to hormones is likely to contribute to early subclinical stages of mammary tumorigenesis, resulting in additional mammary tumors in the subset of dogs that remained sexually intact.
Relatively few (n = 9) dogs died because of new mammary tumors, and all of them occurred late relative to the initial tumor surgery. Consequently, the increased rate of new tumors in the intact group did not translate into a significant difference in overall survival between the 2 treatment groups. In addition to the increased risk of new mammary tumors, nine of the intact dogs also experienced significant uterine and ovarian diseases necessitating emergency surgery or even euthanasia in some cases.
In this study, cause of death was only attributed to mammary tumors if the presence of mammary tumors and metastasis was the direct cause of death/euthanasia. The majority of the dogs in this study were euthanized because of other causes; however, many (n = 32) of these dogs did have documented new mammary tumors at the time of euthanasia. Because of the complexities of issues affecting decisions regarding euthanasia, survival may represent a “soft” and biased endpoint in this study. In contrast, the development of new tumors represents an objective and clearly defined endpoint and our results show that OHE significantly decreases the risk for new tumors in dogs with NMT.
Alternatively or in addition to OHE, prophylactic mastectomies of normal mammary glands could be considered to prevent new mammary tumors. To make good decisions regarding which glands to prophylactically remove, the likely localization of such new tumors must be known. According to a previous study, dogs with 1 mammary tumor were likely to develop new tumors ipsilaterally. The majority of dogs with mammary tumors, however, have more than 1 tumor. In this study, those dogs were also included when determining the localization of new tumors. Our results showed that most new tumors developed adjacent to a previously affected gland; however, almost half of the dogs (n = 5/12) with a single mammary tumor developed a new tumor in the contralateral chain. These findings are in conflict with Stratmann et al's findings, and reflect that the exact localization of new tumors may be somewhat unpredictable. Clearly, a bilateral radical mastectomy would effectively prevent any new mammary tumors in dogs. This is, however, an aggressive procedure and may be difficult to justify as 36% of the intact dogs and 64% of the spayed dogs did not develop additional mammary tumors. In contrast, most dogs recover quickly from OHE. Furthermore, the benefits are 2-fold: OHE significantly reduces the risk for new tumors and prevents uterine/ovarian diseases later on. As many as nine of the 42 dogs (21%) in the nonOHE group developed uterine–ovarian diseases, of which four were euthanized. The previously published data regarding the incidence of pyometra in dogs with mammary tumors are unclear. Some studies suggest that a significant percentage of bitches with mammary tumors have concomitant uterine–ovarian disease, and that they are likely to experience clinical signs (pyometra, mucometra) in the future if OHE is not performed. Others, however, did not find any increased risk of pyometra in dogs with mammary tumors. In this study, 6 of 42 dogs (14%) developed pyometra.
Nonmalignant mammary tumors are generally associated with a long postsurgical survival, thus allowing new tumors and other diseases to develop. As age is advancing, the risk of uterine diseases increases.[26, 27] Some owners choose not to do more surgery in an old dog if it already has been put through 1 or more surgeries earlier. Otherwise treatable uterine and ovarian diseases, may for this reason lead to the decision to euthanize the dog rather than “putting an older dog through another surgery”. OHE at the time of tumor removal may decrease the risk of new tumors, but also protect the dog from later uterine/ovarian diseases which again might save some dogs from being euthanized because of the negative influence such concurrent diseases might have on the owner's decision to pursue additional surgery in an old dog. Our study shows that OHE confers a double benefit in these dogs.
This study's main strength lies in the design. Prospective randomized clinical trials are considered the most reliable form of acquiring scientific evidence because they reduce spurious causality and bias by ensuring comparable groups. Thus, any significant differences between groups in the outcome event can be attributed to the intervention (OHE) and not to some other unidentified factor. Unfortunately, this study was not published in time to be included in the meta-analysis on the effect of OHE on mammary tumor risk performed by Beauvais et al (2012). In fact, this study is the first investigating this question using this design. Additional strength of this study is the long follow-up (median 31.5 months, range 3.5–87.5). Few studies include more than 2 years of follow-up.
The main limitation of this study is that the postsurgical information for some dogs was retrieved solely through phone (17%). However, most of these owners had detected the initial tumors by themselves, thus confirming their ability to recognize such tumors. Moreover, we were only able to obtain the histopathological tumor diagnosis in 26 of the 42 dogs with new tumors. The fact that only 33 dogs were necropsied is also a limitation, but owner contact through phone ensured reliable information regarding why the dog was euthanized. Furthermore, these limitations are not likely to introduce bias between the groups and thus affect the outcome as these sources of bias were evenly distributed between the 2 intervention groups. Finally, the small sample size is a potential weakness when considering multiple variables because the power of detecting effect that really are present decreases with increasing number of variables. However, the significant effect of the OHE-status variable was maintained in the multivariable analysis, indicating that the power of detecting an effect of this variable was indeed acceptable.
In conclusion, OHE at the time of tumor removal reduced the risk for subsequent mammary tumors by 47% in bitches with NMT in this study. This protective effect was evident from 5 months after the OHE intervention and increased until a steady state was reached after approximately 4 years. No significant effect of OHE on overall or tumor-specific survival was detected for this sample of dogs. Adjunctive OHE should be considered in the treatment of adult dogs with mammary tumors.
The authors thank all the dog owners who participated with their dog in the study and gratitude is given to Randi Krontveit for assistance with the survival analysis.
This work was supported by Morris Animal Foundation, the Norwegian “Research foundation for canine cancer”, the Scientific Foundation from the Norwegian Small Animal Veterinary Association, and the Norwegian School of Veterinary Science.
Conflict of Interest Declaration: Authors disclose no conflict of interest.
Stata 11; Stata Corporation, College Station, TX