search

chjackson / flexsurv

The flexsurv R package for flexible parametric survival and multi-state modelling
http://chjackson.github.io/flexsurv/
55 stars 27 forks source link

Flexsurvspline: left and right censored data #9

Closed misstat closed 2 years ago

misstat commented 8 years ago

Fitting a model of the type:

m1<-flexsurvspline(Surv(Age_Recruitment,Age_Exit,dead)~variable1,data=data,k=1,scale="hazard")

I can see the model doesn't fit using this type of Surv object, with right and left censored data, however this running very well inside survival package (this also happens for k greater than 1). The error message says:

Error in solve.QP(Dmat = t(Xq) %*% Xq, dvec = t(t(y) %*% Xq), Amat = t(dXq),  : 
matrix D in quadratic function is not positive definite!

However if I fit the model in this way:

m2<-flexsurvspline(Surv(time_following,dead)~variable1,data=data,k=1,scale="hazard")

convergence it is possible. Obviously I am more interested in fit the first model. Why this error occurs ?

Best regards!

chjackson commented 8 years ago

Thanks for the report. Could I have a minimal reproducible example please so I can fix?

The error happens when trying to find initial values for the spline model. The solve.QP function is used to find values that satisfy the constraint that the cumulative hazard has to be increasing. So you could work around this bug by supplying your own initial values if you can. There's also a backup routine internal to the package that I could make it use in cases where the QP method fails.

On 23/02/16 09:48, misstat wrote:

Fitting a model of the type:

m1<-flexsurvspline(Surv(Age_Recruitment,Age_Exit,dead)~variable1,data=data,k=1,scale="hazard")

I can see the model doesn't fit using this type of Surv object, with right and left censored data, however this running very well inside survival package (this also happens for k greater than 1). The error message says:

|Error in solve.QP(Dmat = t(Xq) %% Xq, dvec = t(t(y) %% Xq), Amat = t(dXq), : matrix D in quadratic function is not positive definite! |

However if I fit the model in this way:

m2<-flexsurvspline(Surv(time_following,dead)~variable1,data=data,k=1,scale="hazard")

convergence it is possible. Obviously I am more interested in fit the first model. Why this error occurs ?

Best regards!

— Reply to this email directly or view it on GitHub https://github.com/chjackson/flexsurv-dev/issues/9.

misstat commented 8 years ago

Thank you for your quick response. I will try to put inits for convergence. And also, please accept my apologies because I am working with confidential data and I can not send a reproducible example. ps: Thanks for the package!

chjackson commented 8 years ago

The latest commit works around cases where this matrix is not positive definite, but I'm still not clear when this can happen.