Flexible parametric modelling of cause-specific hazards to estimate cumulative incidence functions
- Sally R Hinchliffe^{1, 3}Email author and
- Paul C Lambert^{1, 2}
DOI: 10.1186/1471-2288-13-13
© Hinchliffe and Lambert; licensee BioMed Central Ltd. 2013
Received: 16 August 2012
Accepted: 28 January 2013
Published: 6 February 2013
Abstract
Background
Competing risks are a common occurrence in survival analysis. They arise when a patient is at risk of more than one mutually exclusive event, such as death from different causes, and the occurrence of one of these may prevent any other event from ever happening.
Methods
There are two main approaches to modelling competing risks: the first is to model the cause-specific hazards and transform these to the cumulative incidence function; the second is to model directly on a transformation of the cumulative incidence function. We focus on the first approach in this paper. This paper advocates the use of the flexible parametric survival model in this competing risk framework.
Results
An illustrative example on the survival of breast cancer patients has shown that the flexible parametric proportional hazards model has almost perfect agreement with the Cox proportional hazards model. However, the large epidemiological data set used here shows clear evidence of non-proportional hazards. The flexible parametric model is able to adequately account for these through the incorporation of time-dependent effects.
Conclusion
A key advantage of using this approach is that smooth estimates of both the cause-specific hazard rates and the cumulative incidence functions can be obtained. It is also relatively easy to incorporate time-dependent effects which are commonly seen in epidemiological studies.
Keywords
Competing risks Flexible parametric model Cause-specific hazardsBackground
In epidemiological studies two main measures of interest are the risk of an event occurring (probability) and the rate at which it occurs (hazard) [1]. Patients will often be at risk from more than one mutually exclusive event and the occurrence of one of these may alter or prevent the probability of any other event occurring [2]. In this paper we focus on situations where the events are deaths from different causes and so it follows that any event will prevent the others from occurring. In this competing risks scenario, the cause-specific hazard will give the cause-specific mortality rate and the cumulative incidence function will give the proportion of patients at any one time that have died from a particular cause [3].
There are two main approaches to modelling competing risks [4]. The first is to model the cause-specific hazards and transform these to obtain the cumulative incidence function. The second is to model the cumulative incidence function directly [5]. We advocate the first approach as both the cause-specific hazards and the cumulative incidence function can provide a better understanding of risk factors and their effect on the population as a whole [1]. Cause--specific hazards can inform us about the impact of risk factors on rates of disease or mortality, while the cumulative incidence functions provide an absolute measure with which to base prognosis and clinical decisions on [6].
Competing risks analyses are being increasingly carried out in epidemiological studies. However, the methodology applied varies and is not always optimal. Often, separate analyses will be carried out for each competing event and only the cause-specific hazard ratios will be reported for each [7–9]. This method is not wrong if the researchers are only interested in the rate of disease or mortality. However, without estimating an absolute measure such as the cumulative incidence function, it is difficult to communicate these results in terms of the impact that risk factors have at a population level. In comparison, other researchers choose to model on the cumulative incidence scale using the Fine and Gray method and, therefore, provide no information on the cause-specific hazards [10, 11].
In many research papers, the model used to estimate the cause-specific hazards will be different from the model used to estimate the cumulative incidence functions. For example, the cause-specific hazard ratios are reported from a Cox proportional hazards regression model but the cumulative incidence functions are estimated non-parametrically and separately for different subgroups of patient [12–14]. Whilst non-parametric approaches are good for describing the data, there are many advantages for the use of modelling techniques in observational studies when there are a number of covariates that need to be adjusted for.
Many regression models used to estimate cumulative incidence functions will assume proportional hazards. In large epidemiological studies the assumption of proportional hazards is often unreasonable. Therefore, a model that can easily incorporate time-dependent effects is desirable.
In summary, we would like to be able to model competing risks scenarios using the approach that estimates both the cause-specific hazards and the cumulative incidence functions as we believe both to be useful measures. We would like to obtain smooth estimates for both of these measures rather than considering a step function. Finally, we want to be able to incorporate time-dependent effects for one or all of the competing events. Whilst the majority of the above can be addressed within a Cox modelling framework, we feel that parametric models have the advantage of directly estimating cause-specific hazard rates in the model as well as handling non-proportional hazards with ease. For these reasons, we advocate the use of the flexible parametric survival model to obtain both the cause-specific hazards and the cumulative incidence function in a competing risks framework.
Methods
Competing risks
where h _{ k,0 }(t) is the baseline cause-specific hazard for cause k and β _{ k } are the covariate effects (log hazard ratios).
Under the assumption that the competing events are independent (conditional on covariates), the complement of the cause-specific survival function can be interpreted as the probability of dying from cause k in a hypothetical world where it is not possible to die from anything else [15]. In many situations the assumption of independence will not be reasonable in which case any estimates obtained through Equation (3) are not interpretable as probabilities. Even under the strong assumption of independence, these estimates of cause-specific survival are of little use to patients making decisions in the real world where death from other causes play a big role. Therefore, a better approach may be to acknowledge that patients may die from something else other than their cancer.
The cumulative incidence function is not only a function of the cause-specific hazard for the event of interest but also incorporates the cause-specific hazards for the competing events [1]. Previous research has mainly focussed on the use of the Cox model or non-parametric estimates in a competing risks framework [16, 17]. Here, we advocate the use of the flexible parametric model.
Flexible parametric model
This is now a linear function of log-time. However, rather than assuming linearity with ln(t) the flexible parametric model uses restricted cubic splines for ln(t) [19]. The log cumulative hazard function is used as opposed to the hazard function as the “end artefacts” in the fitted spline functions at the extremes of the time scale are more severe for the hazard function. Furthermore, implementing on the log time scale means that the fitted function is typically gently curved or nearly linear, and is usually very smooth [18]. Finally, modelling on this scale means it is easy to transform to the survival and hazard functions [20].
Regression splines are piecewise polynomial functions that are forced to join at predefined points on the x-axis. These joining points are known as knots. In order to obtain a smooth function the regression splines are also forced to have continuous first and second derivatives. For restricted cubic splines a further restriction forces the splines to be linear beyond the boundary knots.
and (u)_{+} = u if u > 0 and 0 if u ≤ 0. Thus, a model with N knots for the baseline log cumulative hazard uses N-1 degrees of freedom.
The number of spline variables for a particular time-dependent effect will depend on the number of knots, n _{ j } [15].
As shown in Equation (4), the cumulative incidence is a function of the cause-specific hazard functions. The cause-specific hazard function can be obtained from the flexible parametric model through Equation (13) by only considering one cause of death at a time and censoring competing events. Alternatively, we can stack the data and fit one model for all K causes simultaneously. This approach is described in further detail later in the paper.
The integral in Equation (4) can be obtained numerically. The integration is performed using similar methods to those proposed by Carstensen [22] and Lambert et al. [15]. The formulae for these methods are given in Appendix 1. It is possible to construct confidence intervals for the cumulative incidence function under the Cox model [23]. However, this is by no means a trivial task [23, 24]. An advantage of our approach is that confidence intervals can be obtained using the delta method as the baseline hazards are estimated as part of the model (see Appendix 1).
Two user-friendly commands have been written in Stata that implement the methodology described in this paper. The command stpm2 will fit a flexible parametric survival model [21] and the command stpm2cif can be used to obtain the cumulative incidence functions through post-estimation [25]. Example code for these commands can be found in Appendix 2.
Relative measures
This can be interpreted as the probability of having died from cause k given that a death has occurred by time t.
This can be interpreted as the probability of having died from cause k given that a death has occurred at time t.
Illustrative example
One research area that is increasingly making use of competing risks methodology is population based cancer studies. Here we use data obtained from the SEER public use dataset [26] on survival of breast cancer patients. The patients analysed were all white females aged between 18 and 103 and were diagnosed between the years 1996 and 2005. Patients that were diagnosed at death or autopsy (n = 509) or had an unknown cause of death (n = 546) were excluded from the analyses. Only patients with a first primary malignant indicator were included (n = 18,433 excluded). If the stage of breast cancer was unknown then the patient was also excluded (n = 991). This left a total of 38,544 patients to be analysed.
Number (%) of patients in each age group and stage of breast cancer at diagnosis
Age group | Localised | Regional | Distant | Total |
---|---|---|---|---|
18-59 | 10,712 (55.6) | 7,467 (38.8) | 1,084 (5.6) | 19,263 (100) |
60-69 | 5,249 (64.3) | 2,414 (29.6) | 490 (6.1) | 8,153 (100) |
70-79 | 4,884 (68.1) | 1,884 (26.2) | 411 (5.7) | 7,179 (100) |
80+ | 2,645 (67) | 983 (24.9) | 321 (8.1) | 3, 949 (100) |
Total | 23,490 | 12,748 | 2,306 | 38,544 |
Expanding the data set
ID | Age | Time | Cause | Status |
---|---|---|---|---|
1 | 50 | 10 | Breast Cancer | 0 |
1 | 50 | 10 | Other Cancer | 0 |
1 | 50 | 10 | Heart Disease | 0 |
1 | 50 | 10 | Other Causes | 0 |
2 | 70 | 6.5 | Breast Cancer | 0 |
2 | 70 | 6.5 | Other Cancer | 0 |
2 | 70 | 6.5 | Heart Disease | 1 |
2 | 70 | 6.5 | Other Causes | 0 |
Results and discussion
Proportional hazards models
Both a Cox-proportional hazards model and a flexible parametric proportional hazards model were fitted in order to make a comparison of the two models in terms of both the cause-specific hazard ratios and the cumulative incidence function. The Cox proportional hazards model does not directly estimate the baseline hazard, h_{k,0}(t), therefore, when obtaining the cumulative incidence functions the Breslow method for the cumulative baseline hazard needs to be substituted into Equation (4). However, if the cause-specific hazard rates were required then the baseline hazards would need to be estimates through post-estimation using, for example, kernel smoothing [21]. For the flexible parametric model the baseline knots were positioned differently for each of the four causes. The knot locations were chosen by taking the first and last event times along with the 25^{th}, 50^{th} and 75^{th} centiles of the event times for each of the four causes.
As shown in Table 2, the data has been stacked so that each patient now has four rows of data, one for each cause. If the effects of age and stage were believed to be the same for each of the four causes of death then the stacked data format would allow us to share the parameters across all four causes. However, in this example, the effects of both age group and stage at diagnosis are different for each cause. We could revert to fitting a separate model for each of the four causes of death but for demonstrative purposes we have instead fitted interaction terms between each cause and each of the two variables. Further details of this can be seen in the Stata code in Appendix 2.
Comparison of Cox proportional hazards model (Cox) and flexible parametric proportional hazards model (FPM), hazard ratios (95% confidence intervals)
Breast cancer | Other cancer | |||
---|---|---|---|---|
Cox | FPM | Cox | FPM | |
Ages 18-59 | 1.00 | 1.00 | 1.00 | 1.00 |
Ages 60-69 | 0.90 (0.83, 0.97) | 0.90 (0.83, 0.98) | 2.12 (1.52, 2.94) | 2.12 (1.52, 2.95) |
Ages 70-79 | 1.27 (1.17, 1.37) | 1.27 (1.17, 1.37) | 3.18 (2.31, 4.37) | 3.19 (2.32, 4.38) |
Ages 80+ | 2.08 (1.90, 2.28) | 2.09 (1.91, 2.29) | 6.59 (4.73, 9.17) | 6.63 (4.76, 9.23) |
Localised | 1.00 | 1.00 | 1.00 | 1.00 |
Regional | 4.15 (3.85, 4.47) | 4.15 (3.85, 4.47) | 2.15 (1.61, 2.88) | 2.16 (1.61, 2.88) |
Distant | 33.68 (31.08, 36.50) | 33.84 (31.23, 36.67) | 25.58 (19.18, 34.12) | 25.82 (19.36, 34.44) |
Heart disease | Other causes | |||
Cox | FPM | Cox | FPM | |
Ages 18-59 | 1.00 | 1.00 | 1.00 | 1.00 |
Ages 60-69 | 4.76 (3.62, 6.24) | 4.76 (3.62, 6.24) | 3.46 (2.89, 4.14) | 3.46 (2.89, 4.14) |
Ages 70-79 | 17.05 (13.42, 21.67) | 17.07 (13.43, 21.69) | 10.22 (8.73, 11.96) | 10.22 (8.73, 11.96) |
Ages 80+ | 70.57 (55.84, 89.17) | 70.75 (55.99, 89.40) | 31.54 (27.00, 36.84) | 31.60 (27.07, 36.91) |
Localised | 1.00 | 1.00 | 1.00 | 1.00 |
Regional | 1.42 (1.27, 1.60) | 1.42 (1.27, 1.60) | 1.11 (1.01, 1.26) | 1.11 (1.02, 1.22) |
Distant | 2.44 (1.89, 3.14) | 2.46 (1.91, 3.16) | 2.08 (1.67, 2.58) | 2.09 (1.68, 2.60) |
Previous studies have shown a relationship between radiation therapy and cardiovascular mortality [27–29] and a similar relationship for chemotherapy [30]. The likelihood of receiving either radiotherapy or chemotherapy as a treatment for breast cancer increases with the severity of the staging. This could again explain the increased risk of death from heart disease with increasing severity of breast cancer staging [31].
Figure 2 illustrates how the proportional hazard assumption forces the log hazard functions for the three stages to be parallel to each other. We can relax this assumption by incorporating time-dependent effects in the model.
Time-dependent models
For the remaining analyses we only considered a flexible parametric non-proportional hazards model. This model included time-dependent effects for age groups 60–69, 70–79 and 80+ for breast cancer and other causes and also for regional and distant stages for breast cancer, other cancer and other causes. These were selected using likelihood ratio tests (p-value < 0.05). All the time-dependent effects were fitted using 4 degrees of freedom and had the same knot locations as those used in the proportional hazards model.
Relative measures
Confidence intervals
Sensitivity to number of knots
Models with varying degrees of freedom for the baseline time-dependent effects, df _{ b } and the additional time-dependent effects, df _{ t }
Baselinedf _{ b } | Time-dependentdf _{ t } | AIC | BIC | |
---|---|---|---|---|
Model 1 | 4 | 4 | 61841.19 | 62459.84 |
Model 2 | 5 | 5 | 61945.39 | 62606.23 |
Model 3 | 5 | 3 | 61963.30 | 62483.53 |
Model 4 | 7 | 3 | 61947.53 | 61783.53 |
Model 5 | 7 | 4 | 61938.33 | 62585.10 |
Model 6 | 3 | 3 | 61962.75 | 62426.74 |
Conclusions
We have shown how to estimate both the cause-specific hazards and the cumulative incidence functions using a flexible parametric survival model. This approach provides smooth estimates of the cause-specific hazard and the cumulative incidence function, both of which we consider to be measures of interest. The flexible parametric model can easily incorporate time-dependent effects for one or more of the competing events. We have also illustrated two other useful measures that can be obtained with some simple manipulation of the cause-specific hazard and cumulative incidence estimates.
The flexible parametric proportional hazards model produces very similar estimates to the Cox proportional hazards model in terms of both the cause-specific hazard ratios and the cumulative incidence functions. A further alternative is to use a mixture model for competing risks data as proposed by Larson and Dinse [4, 33]. However, this approach has two main disadvantages: it is time consuming and the estimated distribution will depend on the length of follow-up [34].
The confidence intervals obtained through the delta method have been shown to be very similar to those obtained through bootstrapping but have the added advantage of taking considerably less time to compute.
The assumption of proportional hazards is often unreasonable in epidemiological studies. It is important to understand the changing effect of a covariate over the time period rather than just assuming a constant hazard. For example, a treatment may have a large impact on mortality early on in the follow-up period but this effect could diminish as time goes on [35]. It is, therefore, important to consider methods such as those described in this paper, that can account for time-dependent effects. The flexible parametric model may be criticized as the number and location of the knots are subjective. However, the sensitivity analysis demonstrates that the knot location has very little impact in terms of the cumulative incidence function. Similar results have been reported elsewhere in relation to the sensitivity of the knots [15, 18, 20, 36].
In this paper we have grouped age into four categories for simplicity whilst illustrating the method. However, it may be preferable to model age continuously using regression splines as has been done in previous papers [37, 38].
The main advantages of the flexible parametric model are in large studies where time-dependent effects will often play a prominent role. In much smaller studies where there are fewer events there may not always be sufficient information to adequately estimate the underlying hazard using this model.
This paper describes modelling cause-specific hazards and using these to obtain the cumulative incidence function. Alternatively, the cumulative incidence function can be modelled directly using, for example, Fine and Grays subdistribution approach [5]. This may be useful when interest only lies in obtaining estimates of the cumulative incidence function for one of the competing events. However, if interest lies in visualising the overall probability broken down by specific events, such as those shown in Figure 2, then it should be noted that the direct regression approach does not have a boundary condition and so in some cases the overall probability may exceed one. We believe that the cause-specific approach, as described here, is advantageous for a full understanding of risk factors and real world implications.
Unlike measures of net survival, the cumulative incidence function allows us to present “real world” probabilities where a patient is not only at risk of dying from their cancer but also from any other cause of death. We can also estimate these “real world” probabilities using relative survival [15]. The advantage of the cause-specific approach is that we can examine more causes of death but this is at the expense of having to rely on cause of death information.
Finally, a user friendly program has been written in Stata to enable users to implement the methodology described in this paper. This command is called stpm2cif and is available from the Statistical Software Components (SSC) archive [25, 39].
Appendix 1–Details of the delta-method used to calculate confidence intervals
- 1.
The time scale is split into a large number, m, of small intervals.
- 2.
The integrand of the cumulative incidence function, $\widehat{f}\left({t}_{m}|{\mathbf{x}}_{\mathbf{0}}\right)$, is predicted for a particular covariate vector, x _{ 0 } at each of the m time intervals, t _{ m }.
- 3.The variance-covariance matrix for the integrand $\widehat{f}\left({t}_{m}|{\mathbf{x}}_{\mathbf{0}}\right)$, is obtained at each time interval using the delta method. The Stata command predictnl calculates the observation-specific derivatives for each parameter in the model. If we let G be the m × p matrix of observation-specific derivatives then the variance-covariance matrix can be estimated using the equation$\mathit{Var}\left(\widehat{f}\left({t}_{m}\right)\right)=G\widehat{V}G$
where $\widehat{V}$ is the estimated variance matrix for the model parameters.
- 4.The cumulative incidence function can then be calculated by summing the values of the integrand for the m time intervals. In order to do this, a triangular matrix L needs to be created. For example, for three intervals this looks like${C}_{k}\left(t\right)=l\times \left[\begin{array}{rrr}\hfill 1& \hfill 0& \hfill 0\\ \hfill 1& \hfill 1& \hfill 0\\ \hfill 1& \hfill 1& \hfill 1\end{array}\right]\phantom{\rule{1em}{0ex}}\left[\begin{array}{c}\hfill \widehat{f}\left({t}_{1}\right)\hfill \\ \hfill \widehat{f}\left({t}_{2}\right)\hfill \\ \hfill \widehat{f}\left({t}_{3}\right)\hfill \end{array}\right]\phantom{\rule{0.5em}{0ex}}=L\phantom{\rule{0.5em}{0ex}}\left[\begin{array}{c}\hfill \widehat{f}\left({t}_{1}\right)\hfill \\ \hfill \widehat{f}\left({t}_{2}\right)\hfill \\ \hfill \widehat{f}\left({t}_{3}\right)\hfill \end{array}\right]$
where l is the interval length.
- 5.The variance-covariance matrix for the cumulative incidence function of the k ^{ th } cause is then calculated using$\mathit{Var}\phantom{\rule{0.5em}{0ex}}({C}_{k}\left(t\right)=\mathit{LG}\widehat{V}{G}^{\prime}{L}^{\prime}$
Appendix 2–Stata analysis code for flexible parametric model section of illustrative example. For more information see the Stata help file [38] or the Stata Journal article [30]
***Expand the data so that each patient has 4 rows – one for each cause of death***
expand 4
bysort id: gen cause = _n
***Generate indicator variables for each cause of death along with an overall indicator ***
gen breast = cause==1
gen cancer = cause==2
gen heart = cause==3
gen other = cause==4
gen event = (cause==cod)
***Create interactions between age group and causes***
gen agebreast = agegrp*breast
gen agecancer = agegrp*cancer
gen ageheart = agegrp*heart
gen ageother = agegrp*other
***Create dummy variables for each age cause interaction***
tab agebreast, gen(agebreast)
tab agecancer, gen(agecancer)
tab ageheart, gen(ageheart)
tab ageother, gen(ageother)
***Re-name age cause dummy variables ***
foreach var in breast cancer heart other {
rename age`var'2 age`var'1
rename age`var'3 age`var'2
rename age`var'4 age`var'3
rename age`var'5 age`var'4
}
*** Create interactions between stage and causes***
gen stagebreast = seerhistoricstage*breast
gen stagecancer = seerhistoricstage*cancer
gen stageheart = seerhistoricstage*heart
gen stageother = seerhistoricstage*other
***Create dummy variables for each stage cause interaction***
tab stagebreast, gen(stagebreast)
tab stagecancer, gen(stagecancer)
tab stageheart, gen(stageheart)
tab stageother, gen(stageother)
*** Re-name stage cause dummy variables ***
foreach var in breast cancer heart other {
rename stage`var'2 stage`var'1
rename stage`var'3 stage`var'2
rename stage`var'4 stage`var'3
}
***stset the data to tell Stata we are dealing with survival data***
stset exit, origin(dx) failure(event) scale(365.24) exit(time dx + (10*365.24))
*** Fit a flexible parametric proportional hazards model using stpm2 command***
stpm2 breast cancer heart other agebreast? agecancer? ageheart? ageother? ///
stagebreast? stagecancer? stageheart? stageother?, ///
scale(hazard) rcsbaseoff nocons ///
tvc(breast cancer heart other) initstrata(cause) ///
knotstvc(breast 1.37 2.62 4.70 ///
cancer 1.00 2.95 5.87 ///
heart 1.79 3.87 6.37 ///
other 1.95 3.95 6.46) ///
bknotstvc(breast 0.038 9.96 ///
cancer 0.04 9.96 ///
heart 0.04 9.96 ///
other 0.04 9.96)
***Predict the cumulative incidence functions, the cause-specific hazard rates, the contribution to the total mortality and the contribution to the overall hazard for each covariate pattern using stpm2cif command***
forvalues j = 1/4 {forvalues l = 1/3 {
if `j'! = 1 {
}
if `j'==1 {
}
}
if `l'==1 {
stpm2cif breast`j'`l' cancer`j'`l' heart`j'`l' other`j'`l', ///
cause1(breast 1 agebreast`j' 1) ///
cause2(cancer 1 agecancer`j' 1) ///
cause3(heart 1 ageheart`j' 1) ///
cause4(other 1 ageother`j' 1) haz conthaz contmort
}
if `l'! = 1 {
stpm2cif breast`j'`l' cancer`j'`l' heart`j'`l' other`j'`l', ///
cause1(breast 1 agebreast`j' 1 stagebreast`l' 1) ///
cause2(cancer 1 agecancer`j' 1 stagecancer`l' 1) ///
cause3(heart 1 ageheart`j' 1 stageheart`l' 1) ///
cause4(other 1 ageother`j' 1 stageother`l' 1) haz conthaz contmort
}
if `l'==1 {
stpm2cif breast`j'`l' cancer`j'`l' heart`j'`l' other`j'`l', ///
cause1(breast 1) ///
cause2(cancer 1) ///
cause3(heart 1) ///
cause4(other 1) haz conthaz contmort
}
if `l'! = 1 {
stpm2cif breast`j'`l' cancer`j'`l' heart`j'`l' other`j'`l', ///
cause1(breast 1 stagebreast`l' 1) ///
cause2(cancer 1 stagecancer`l' 1) ///
cause3(heart 1 stageheart`l' 1) ///
cause4(other 1 stageother`l' 1) haz conthaz contmort
}
}
Declarations
Authors’ Affiliations
References
- Andersen PK, Geskus RB, Witte T, Putter H: Competing risks in epidemiology: possibilities and pitfalls. Int J Epidemiol. 2012, 41 (3): 861-870. 10.1093/ije/dyr213.View ArticlePubMedPubMed CentralGoogle Scholar
- Geskus RB: Cause-specific cumulative incidence estimation and the Fine and Gray model under both left truncation and right censoring. Biometrics. 2011, 67: 39-49. 10.1111/j.1541-0420.2010.01420.x.View ArticlePubMedGoogle Scholar
- Prentice RL, Kalbfleisch JD, Peterson AV, Flournoy N, Farewell VT, Breslow NE: The analysis of failure times in the presence of competing risks. Biometrics. 1978, 34: 541-554. 10.2307/2530374.View ArticlePubMedGoogle Scholar
- Lau B, Cole SR, Gange SJ: Competing risk regression models for epidemiologic data. Am J Epidemiol. 2009, 170: 244-256. 10.1093/aje/kwp107.View ArticlePubMedPubMed CentralGoogle Scholar
- Fine JP, Gray RJ: A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999, 94: 496-509. 10.1080/01621459.1999.10474144.View ArticleGoogle Scholar
- Koller MT, Raatz H, Steyerberg EW, Wolbers M: Competing risks and the clinical community: irrelevance or ignorance?. Stat Med. 2011, 31: 1089-1097.View ArticlePubMedPubMed CentralGoogle Scholar
- Colzani E, Liljegren A, Johansson ALV, Adolfsson J, Hellborg H, Hall PFL: Prognosis of Patients With Breast Cancer: Causes of Death and Effects of Time Since Diagnosis, Age, and Tumor Characteristics. J Clin Oncol. 2011, 29: 4014-4021. 10.1200/JCO.2010.32.6462.View ArticlePubMedGoogle Scholar
- Baer HJ, Glynn RJ, Hu FB, Hankinson SE, Willett WC, Colditz GA: Risk factors for mortality in the Nurses' Health Study: a competing risks analysis. Am J Epidemiol. 2011, 173: 319-329. 10.1093/aje/kwq368.View ArticlePubMedGoogle Scholar
- Pocobelli G, Peters U, Kristal AR, White E: Use of supplements of multivitamins, vitamin C, and vitamin E in relation to mortality. Am J Epidemiol. 2009, 170: 472-483. 10.1093/aje/kwp167.View ArticlePubMedPubMed CentralGoogle Scholar
- Kutikov A, Egleston BL, Wong Y-N, Uzzo RG: Evaluating overall survival and competing risks of death in patients with localized renal cell carcinoma using a comprehensive nomogra. J Clin Oncol. 2010, 28: 311-317. 10.1200/JCO.2009.22.4816.View ArticlePubMedGoogle Scholar
- Pestalozzi BC, Zahrieh D, Price KN, Holmberg SB, Lindtner J, Collins J: Identifying breast cancer patients at risk for Central Nervous System (CNS) metastases in trials of the International Breast Cancer Study Group (IBCSG). Ann Oncol. 2006, 17: 935-944. 10.1093/annonc/mdl064.View ArticlePubMedGoogle Scholar
- De Bruin ML, Sparidans J, Veer MB, Noordijk EM, Louwman MWJ, Zijlstra JM: Breast cancer risk in female survivors of Hodgkin's Lymphoma: lower risk after smaller radiation volumes. J Clin Oncol. 2009, 27: 4239-4246. 10.1200/JCO.2008.19.9174.View ArticlePubMedGoogle Scholar
- Glynn RJ, Rosner B: Comparison of risk factors for the competing risks of coronary heart disease, stroke, and venous thromboembolism. Am J Epidemiol. 2005, 162: 975-982. 10.1093/aje/kwi309.View ArticlePubMedGoogle Scholar
- Simard EP, Pfeiffer RM, Engels EA: Cumulative incidence of cancer among individuals with acquired immunodeficiency syndrome in the United States. Cancer. 2011, 117: 1089-1096. 10.1002/cncr.25547.View ArticlePubMedGoogle Scholar
- Lambert PC, Dickman PW, Nelson CP, Royston P: Estimating the crude probability of death due to cancer and other causes using relative survival models. Stat Med. 2010, 29: 885-895. 10.1002/sim.3762.View ArticlePubMedGoogle Scholar
- Lunn M, McNeil D: Applying Cox regression to competing risks. Biometrics. 1995, 51: 524-532. 10.2307/2532940.View ArticlePubMedGoogle Scholar
- Satagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD: A note on competing risks in survival data analysis. Br J Cancer. 2004, 91: 1229-1235. 10.1038/sj.bjc.6602102.View ArticlePubMedPubMed CentralGoogle Scholar
- Royston P, Parmar MKB: Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects. Stat Med. 2002, 21: 2175-2197. 10.1002/sim.1203.View ArticlePubMedGoogle Scholar
- Durrleman S, Simon R: Flexible regression models with cubic splines. Stat Med. 1989, 8: 551-561. 10.1002/sim.4780080504.View ArticlePubMedGoogle Scholar
- Royston P, Lambert PC: Flexible parametric survival analysis using Stata: beyond the Cox model. 2011, Stata Press booksGoogle Scholar
- Lambert PC, Royston P: Further development of flexible parametric models for survival analysis. Stata J. 2009, 9: 265-290.Google Scholar
- Carstensen B: Technical report. Demography and epidemiology: Practical use of the lexis diagram in the computer age or: Who needs the Cox model anyway?. 2006, University of Copenhagen: Department of BiostatisticsGoogle Scholar
- Cheng SC, Fine JP, Wei LJ: Prediction of cumulative incidence function under the proportional hazards model. Biometrics. 1998, 54: 219-228. 10.2307/2534009.View ArticlePubMedGoogle Scholar
- Braun TM, Yuan Z: Comparing the small sample performance of several variance estimators under competing risks. Stat Med. 2007, 28: 1170-1180.View ArticleGoogle Scholar
- Hinchliffe SR, Lambert PC: Extending the flexible parametric survival model for competing risks. Stata J. 2012, In PressGoogle Scholar
- National Cancer Institute: Surveillance, Epidemiology, and End Results (SEER) Program Research Data (1973–2008). 2011, (http://www.seer.cancer.gov)Google Scholar
- Bouillon K, Haddy N, Delaloge S, Garbay JR, Garsi JP, Brindel P: Long-Term Cardiovascular Mortality After Radiotherapy for Breast Cancer. J Am Coll Cardiol. 2011, 57: 445-452. 10.1016/j.jacc.2010.08.638.View ArticlePubMedGoogle Scholar
- McGale P, Darby SC, Hall P, Adolfsson J, Bengtsson NO, Bennet AM: Incidence of heart disease in 35,000 women treated with radiotherapy for breast cancer in Denmark and Sweden. Radiother Oncol. 2011, 100: 167-175. 10.1016/j.radonc.2011.06.016.View ArticlePubMedGoogle Scholar
- Hooning MJ, Botma A, Aleman BMP, Baaijens MHA, Bartelink H, Klijn JGM: Long-term risk of cardiovascular disease in 10-year survivors of breast cancer. J Natl Cancer Inst. 2007, 99: 365-375. 10.1093/jnci/djk064.View ArticlePubMedGoogle Scholar
- Pinder MC, Duan Z, Goodwin JS, Hortobagyi GN, Giordano SH: Congestive heart failure in older women treated with adjuvant anthracycline chemotherapy for breast cancer. J Clin Oncol. 2007, 25: 3808-3815. 10.1200/JCO.2006.10.4976.View ArticlePubMedGoogle Scholar
- Fang F, Fall K, Mittleman MA, Sparén P, Weimin Ye W, Adami H-O: Suicide and cardiovascular death after a cancer diagnosis. N Engl J Med. 2012, 366: 1310-1318. 10.1056/NEJMoa1110307.View ArticlePubMedGoogle Scholar
- Efron B, Tibshirani RJ: An introduction to the bootstrap. 1993, New York: Chapman & HallView ArticleGoogle Scholar
- Larson MG, Dinse GE: A mixture model for the regression analysis of competing risks data. Appl. Statist. 1985, 34: 201-211. 10.2307/2347464.View ArticleGoogle Scholar
- Nicolaie MA, van HC H, Putter H: Vertical modeling: A pattern mixture approach for competing risks modeling. Stat Med. 2009, 29: 1190-1205.Google Scholar
- Jatoi I, Anderson WF, Jeong J-H, Redmond CK: Breast cancer adjuvant therapy: time to consider its time-dependent effects. J Clin Oncol. 2011, 29: 2301-2304. 10.1200/JCO.2010.32.3550.View ArticlePubMedPubMed CentralGoogle Scholar
- Nelson CP, Lambert PC, Squire IB, Jones DR: Flexible parametric models for relative survival, with application in coronary heart disease. Stat Med. 2007, 26: 5486-5498. 10.1002/sim.3064.View ArticlePubMedGoogle Scholar
- Lambert PC, Holmberg L, Sandin F, Bray F, Linklater KM, Purushotham A: Quantifying differences in breast cancer survival between England and Norway. Cancer Epidemiol. 2011, 35: 536-533.View ArticleGoogle Scholar
- Eloranta S, Lambert PC, Andersson TML, Czene K, Hall P, Björkholm M, Dickman PW: Partitioning of excess mortality in population-based cancer patient survival studies using flexible parametric survival models. BMC Med Res Methodol. 2012, 12: 86-10.1186/1471-2288-12-86.View ArticlePubMedPubMed CentralGoogle Scholar
- Hinchliffe SR, Lambert PC: Statistical Software Components. STPM2CIF: Stata module to estimate cumulative incidence function. 2011, Boston College Department of Economics: Statistical Software ComponentsGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2288/13/13/prepub
Pre-publication history
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.