Dear Statalists,
Here I am again seeking your help!!! I am trying to develop a clinical study to assess the learning curve of a specific surgical procedure in the following cohort:
To do so I want to combine the use of restricted cubic splines and CUSUM method. My outcomes of interest are either continuous (SurgicalTimemin Enucleationtimemin Morcellationtimemin) or dichotomous (Readmission Irritativesymptomsat3months Stressincontinence3months), while ID indicates the number of surgical procedures.
Given the hypothesis that the effect of experience on the outcomes of interest is nonlinear as a result of a learning process, I want to model experience using restricted cubic splines where significant during regression analysis.
I want to use the CUSUM method to have a graphical representation of the learning curve. To do so, for binary outcomes, I want to generate the cumulative log-likelihood CUSUM curve, with control limits determined using a simulation method and based on detecting a 100% increase (odds ratio = 2) or 50% decrease (odds ratio = 0.5) in the odds of experiencing the outcome with a Type-I error rate of 10%.
For continuous outcomes, I want to generate a standard CUSUM curve, with control limits set to detect outcome values 3 standard errors from the target process mean. I also would like to generate the observed minus expected CUSUM curve for both binary and continuous outcomes, and apply LOWESS smoothers to the O-E CUSUM plots to help visualize trends in the curve. The expected or target values for all outcomes will be based on the outcomes achieved with the current gold standard for this procedure.
I have been trying to perform this calculation in different ways, using the commands cusum and cusum6 but I need to apply this also to continuous variables (so the command custom is not correct) and not with a time variable (so cusum6 is not the right one).
It would be great if anyone in the forum could help me set up this project. I am more than willing to include the person who would help as a co-author in the manuscript that will derive from this.
Thank you in advance,
Francesco Ditonno
Here I am again seeking your help!!! I am trying to develop a clinical study to assess the learning curve of a specific surgical procedure in the following cohort:
Code:
* Example generated by -dataex-. For more info, type help dataex clear input int(ID_number SurgicalTimemin Enucleationtimemin Morcellationtimemin) byte(Readmission Irritativesymptomsat3months Stressincontinence3months) 28 31 . . 0 2 1 79 37 25 2 0 1 0 176 37 16 3 0 . . 197 41 30 7 1 . . 192 44 38 4 0 . . 155 44 28 3 1 0 0 235 45 18 13 . . . 226 46 28 8 . . . 175 46 28 10 0 . . 80 49 45 5 1 . . 51 49 . . 0 0 0 158 49 43 2 0 . . 134 52 30 7 0 . . 44 52 . . 0 3 0 128 55 . . 1 0 0 211 55 . . 0 . . 170 55 30 12 0 . . 231 55 35 8 . . . 132 56 35 7 0 . . 47 56 . . 1 1 0 56 57 35 10 0 0 0 34 57 . . 0 0 0 101 58 45 7 0 1 0 217 58 38 9 . . . 157 58 . . 0 0 0 86 58 30 8 0 1 0 93 60 35 7 0 0 0 208 60 40 5 0 . . 203 60 40 10 0 . . 94 61 33 3 0 0 0 105 63 50 5 0 2 1 152 63 47 6 0 . . 42 64 . . 0 2 0 82 65 . . 0 . . 180 65 40 10 0 . . 218 65 35 22 . . . 97 66 52 10 0 0 0 95 66 60 5 0 1 0 53 66 . . 0 0 0 151 66 55 5 1 2 0 222 66 45 8 . . . 146 68 53 6 0 0 0 145 68 60 7 0 1 0 111 69 47 12 0 1 0 78 69 50 10 0 . . 81 69 . . 0 0 0 210 70 60 10 0 . . 227 70 57 16 . . . 182 70 40 8 0 . . 196 70 35 15 0 . . 117 70 56 5 0 0 0 12 71 . . 1 0 0 74 71 60 5 0 0 0 9 72 . . 0 0 0 87 73 59 11 0 0 0 1 73 . . 0 2 1 131 73 63 5 0 2 0 38 74 . . 0 0 0 76 75 . . 0 1 1 68 75 57 7 0 0 0 142 75 60 1 0 1 1 184 75 50 15 0 . . 120 75 23 13 0 1 1 121 75 48 20 0 0 0 55 76 60 7 0 0 0 150 76 52 7 0 . . 24 76 . . 1 0 1 113 76 45 10 0 0 0 41 77 . . 0 1 0 84 79 45 15 0 0 1 160 79 . . 1 2 1 183 80 60 15 0 . . 178 80 35 8 1 . . 206 80 50 10 0 . . 198 80 45 25 0 . . 10 80 . . 1 0 1 187 80 42 12 0 . . 143 80 65 10 0 1 0 69 81 51 10 0 1 1 17 82 . . 0 0 0 163 82 45 24 1 . . 50 83 . . 0 0 0 52 83 . . 0 2 0 107 84 60 6 0 1 0 8 84 . . 0 1 0 141 85 80 20 0 1 0 228 85 70 12 . . . 220 85 70 10 . . . 72 85 60 10 0 2 0 221 85 60 14 . . . 98 87 72 10 0 1 1 125 87 72 8 0 0 0 46 88 . . 1 1 0 14 88 . . 0 0 0 165 90 70 15 0 . . 118 90 . . 0 0 0 169 90 40 10 0 . . 106 90 45 10 0 . . 199 91 72 14 0 . . 37 92 . . 0 1 0 end
Given the hypothesis that the effect of experience on the outcomes of interest is nonlinear as a result of a learning process, I want to model experience using restricted cubic splines where significant during regression analysis.
I want to use the CUSUM method to have a graphical representation of the learning curve. To do so, for binary outcomes, I want to generate the cumulative log-likelihood CUSUM curve, with control limits determined using a simulation method and based on detecting a 100% increase (odds ratio = 2) or 50% decrease (odds ratio = 0.5) in the odds of experiencing the outcome with a Type-I error rate of 10%.
For continuous outcomes, I want to generate a standard CUSUM curve, with control limits set to detect outcome values 3 standard errors from the target process mean. I also would like to generate the observed minus expected CUSUM curve for both binary and continuous outcomes, and apply LOWESS smoothers to the O-E CUSUM plots to help visualize trends in the curve. The expected or target values for all outcomes will be based on the outcomes achieved with the current gold standard for this procedure.
I have been trying to perform this calculation in different ways, using the commands cusum and cusum6 but I need to apply this also to continuous variables (so the command custom is not correct) and not with a time variable (so cusum6 is not the right one).
It would be great if anyone in the forum could help me set up this project. I am more than willing to include the person who would help as a co-author in the manuscript that will derive from this.
Thank you in advance,
Francesco Ditonno
