Meta analysis for studies that applied non-linear*models

[Masoumeh Sanagou]

HI STATALIST,

1-Is there any STATA command to do meta-analysis for studies that applied non-linear models (e.g. linear-exponential) and reported coefficients and their SEs?

2- Does it make sense to do meta-analysis using the studies that reported the results of linear models in combination with those that reported the results of non-linear models?


Kind regards,

[Tiago Pereira]

1. You can use -metan- provided that the coefficients are normally distributed. It is unclear what type of models you have at hand. However, it is relatively frequent that non-linear models result in non-normal distributions for the estimated coefficients.
2. It depends on the models and dependent variables. If the coefficients are estimating different things, then it would make little sense to combine them. It is not possible to provide you with more precise answers, since the details are unclear.

[Masoumeh Sanagou]

Thanks for the reply.

for example we have studies that report:

- Linear-exponential-square-root ERR with model RR = (1+beta*dose)*exp(gamma*sqrt(dose))
- Linear-exponential-squared ERR with model RR = 1+beta*dose*exp(gamma*(dose^2))
- Linear-exponential ERR with model RR = 1+beta*dose*exp(gamma*dose)
- Linear-exponential ERR with model RR = (1+beta*dose)*exp(gamma*dose)



so I can say that because the models estimated the same thing it is make sense to combine them in a meta analysis.

If my understanding is correct, do you have any reference for this?

Kind regards,