Initial identification of a study cohort.

The Mayo Clinic and Olmsted Medical Center institutional review boards approved this study. Written informed consent was given by participants or caregivers in the case of children for their clinical records to be used in this study.

A prior publication described the methods used to create a birth cohort of children born from January 1, 1994 to December 31, 2007 in Olmsted County, MN for the Mayo Anesthesia Safety in Kids (MASK) study [17]. The primary aim of the MASK study is to sample 1,000 children from this birth cohort for detailed prospective neurodevelopmental assessments of behavior and learning to determine associations between anesthesia exposure and these assessments. In this work, the resources of the Rochester Epidemiology Project (REP) [18] and birth certificate information were utilized to identify all children born in Olmsted County, MN over the period of interest. Available birth certificate information included the child’s name, sex, date of birth, the child’s both parent names, and the mother’s address at the time of the child’s birth.

The current manuscript describes how a study cohort was constructed as a subset of the 1994–2007 birth cohort to support the planned study. Although children in both the planned study and the MASK study are drawn from the same birth cohort, they are not necessarily the same children, and the MASK study does not assess the same learning and behavioral outcomes. In particular, the MASK study involves a four-hour detailed neuropsychological assessment to measure multiple cognitive domains. In contrast, the study cohort described in the current paper will be used to examine learning and behavioral outcomes ascertained through school record review, including LD, ADHD, and group-administered achievement tests.

First, children in the birth cohort who were born from January 1, 1996 to December 31, 2000 were identified. This time range was chosen to allow sufficient duration of follow up to identify the outcomes of interest, and to coincide with widespread clinical replacement of the volatile anesthetic halothane with sevoflurane in pediatric practice. The following children were then excluded: 1) children who moved from Olmsted County prior to their third birthday (to ensure that complete information regarding potential exposure to anesthesia and surgery at a young age was available), 2) children who died prior to their 5th birthday, and; 3) children not enrolled in the local school district at age 5 or 6 (from 5th birthday to the day before 7th birthday), as their records would not be available to ascertain outcomes of interest. The following section describes how enrollment status was determined.

Ascertainment of school enrollment status.

A contractual research agreement between the Mayo Clinic and the Independent School District (ISD) 535, which provides public schooling for most children in Olmsted County, MN. This agreement makes accessible for review the cumulative school records of all students living within the boundaries of the school district including those attending any of the parochial or private schools, or were home-schooled within the boundaries of ISD 535. The district provides educational services for all children living within its boundaries evaluated for or requiring any specialized services regardless of which school they attend. For purposes of this paper, these children will be considered to be “enrolled” in ISD 535 (regardless of whether they actually attend ISD 535, a private school or are home schooled). Records are maintained if students move from the school district, graduate, or die. The record includes enrollment records, school assessments, individual and group IQ/achievement test results, special educational intervention, reported long-term outcomes (e.g., drop out, expulsion, high school graduation status, etc.), data and notes related to any type of difficulty in learning, performance, or behavior, as observed by teachers, parents, school psychologists, school nurses, etc., medical records relevant to development, learning or behavior, and correspondence with private tutorial and intervention facilities.

To determine the school enrollment status of birth cohort members, identifiers (including birth certificate information) were merged with the ISD 535 school record files. The merging process involved both automated and manual efforts. The automated process utilized an algorithm developed by the Rochester Epidemiology Project that takes into account “fuzzy” measures of proximity and relative frequency of certain variables such as the last name, first name, date of birth, mother’s name, and address [19]. Cohort children not merged with a school file were manually verified to adjust for misspellings, name and address changes, or other errors. Those children not enrolled in ISD 535 at age 5 or 6 were not eligible for inclusion in the study cohort.

Ascertainment of anesthesia exposures.

The medical records of all children in the birth cohort were reviewed to determine their exposure to general anesthesia prior to the age of 3 years using the Mayo Clinic Surgical Information Retrieval System and a similar resource for procedures performed at Olmsted Medical Center, the only other medical facility within Olmsted County. Details of anesthetic management, including duration, surgical procedure, and drugs utilized, were abstracted.

Propensity matching and the study cohort.

The potential for confounds is inherent in observational studies. In the planned study, the main concern is that exposure to anesthesia and surgery is simply a marker for other factors that primarily contribute to adverse outcomes. To mitigate these potential effects, a propensity matching strategy was used to select children who met other eligibility criteria for inclusion in the study cohort. A similar strategy has been described for selection of children to be invited to participate in the MASK study [17]. Using all members of the birth cohort who were born between January 1, 1996 to December 31, 2000 and who did not meet any of the previously described exclusion criteria, multinomial logistic regression models were used to calculate propensity for receiving general anesthesia. For these models, the dependent variable had 3 levels indicating the number of anesthetic exposures under the age of 3 (none, single, multiple). The explanatory variables included in the model consisted of two sets: 1) 32 binary indicator variables representing the Johns Hopkins ACG Case-Mix System co-morbidity clusters (aggregated diagnostic groups, ADGs) to quantify the overall health status of each cohort child (as described in other prior work [8]); and 2) birth certificate variables including gestational age, birth weight, Apgar scores at 1 and 5 minutes, complications of pregnancy, age of parents, and level of parental education. Because sex has been reported as an important factor associated with LD and ADHD, and we anticipated that the magnitude of confounding for some variables may differ between sexes, separate propensity models were fit for each sex. Since the dependent variable had 3 levels, two propensity scores were calculated: a propensity score for receiving a single exposure and a propensity score for receiving multiple exposures. In order to create sex-specific strata which take into account both propensity scores the following approach was used. For each propensity score, a 5-level nominal variable was created which categorized subjects according to the quintiles of the observed distribution of propensity score. By using all possible combinations of these 2 categorical variables 25 propensity-matched strata were identified for each sex, with the notation (q_single, q_multiple) used to denote these strata. For example, the strata F(1,1) corresponds to females whose propensity score for a single exposure to anesthesia was in the lowest quintile and whose propensity score for multiple exposures to anesthesia was also in the lowest quintile. Similarly, the strata M(1,5) corresponds to males whose propensity score for a single exposure to anesthesia was in the lowest quintile and whose propensity score for multiple exposures to anesthesia was in the highest quintile.

Previous publications suggested an association between multiple exposures to general anesthesia and neurodevelopmental outcomes like LD and ADHD [8, 9, 15]. Therefore the matching strategy aimed to maximize the number of children exposed to general anesthesia in the study cohort, especially those who had multiple anesthetic exposures. Within each of the 50 sex-specific propensity-matched strata, the distribution of children in each exposure (none, single, multiple) group were identified. In strata with at least one child in each of the three exposure groups, all of the exposed individuals (both single and multiple) were selected into the final study cohort along with a random sample of an equal number of children who were not exposed. Using this approach, the goal was to include an equal number of exposed and unexposed children within each stratum. In some strata the number of unexposed children was less than or equal to that of the exposed; for these strata, all available unexposed children were selected into the study cohort. If a given stratum did not contain at least one child from all three exposure groups of children, this stratum was excluded from the study to ensure maximum validity of propensity matching. This was chosen as a conservative approach, but as a result some of the children with anesthesia exposures were not included in the final propensity-matched study cohort.

After the propensity matching had been performed, a measure of socioeconomic status (SES) was calculated for children included in the study cohort and compared among exposure categories. The HOUSES index, developed and validated within the Olmsted County population [20], is calculated by linking address information of the children residence (at age 5 or the age closest to this value) to property data publicly available at local government assessors’ office.

Sample size/power considerations.

To assess whether general anesthesia prior to age 3 years is associated with neurodevelopmental outcomes in the planned study we will compare four outcomes (cumulative frequency of LD and ADHD, number of recipients of an IEP, and scores of group administered achievement tests) between the two groups of children exposed to anesthesia (single or multiple exposures) and those children not exposed to general anesthesia. In a prior study done in a similar birth cohort born from 1976 to 1982, the cumulative frequency of LD prior to age 19 was 36.6%, 23.6%, and 21.3% for children in the multiple, single, and none exposure group, respectively [8]. Assuming that the cumulative frequency of LD is similar, approximately 250 cases of LD would be expected in the current study cohort. With this expected number of cases, the sample size of the present study will provide statistical power of >80% to detect a hazard ratio as low as 1.6 between children with multiple exposures and unexposed children. Similarly, the current study will provide statistical power of >80% to detect a hazard ratio of 1.5 or larger between children with a single exposure compared to those unexposed. For group-administered achievement tests, this sample size will also provide statistical power of 80% to detect a difference of 0.3 standard deviation units between multiply exposed children and children not exposed.

Analysis.

Future reports of the planned study will present the methods used to define and analyze each neurodevelopmental outcome. The present manuscript presents the results of three analyses that characterize the study cohort itself.

The first analysis compared the characteristics of children in the birth cohort who were and were not enrolled in ISD 535 at the age of 5 or 6 years to examine potential bias due to this inclusion requirement for the planned study. Potential reasons that children were not enrolled include death prior to age 5, living or moving out of the boundaries of the school district (small portions of Olmsted County are included in other public school districts), or electing to attend another public school outside of ISD 535 (as allowed by MN state law). Characteristics of children who were and were not enrolled in ISD 535, including birth certificate information, were compared using standard statistical methods. The chi-square test was used for categorical variables and the Wilcoxon rank-sum test was used for ordinal and continuous variables.

The second analysis evaluated the success of the propensity matching schemes in creating groups of children that were similar in measured characteristics (except for anesthesia exposure). In order to take into account the stratified study design, characteristics of the three exposure groups (multiple, single, or none) were compared using Cochran-Mantel-Haenszel tests and stratified generalized linear models were used for categorical and continuous variables respectively.

Finally, since some propensity-matched strata did not contain at least one child from each exposure group, there were some multiply and singly exposed children who were not included in the study cohort. To assess potential biases resulting from this exclusion, the characteristics of children with anesthesia exposures (both single and multiple) who were and were not included in the propensity-matched analysis were compared using the chi-square test for categorical variables and the Wilcoxon rank-sum test for ordinal and continuous variables.