- Email: [email protected]
Decreased health care utilization and health care costs in the inpatient and emergency department setting following initiation of ketogenic diet in pediatric patients: The experience in Ontario, Canada
Accepted Manuscript Title: Decreased Health Care Utilization and Health Care Costs in the Inpatient and Emergency Department Setting Following Initiation of Ketogenic Diet in Pediatric Patients: The Experience in Ontario, Canada Authors: Sharon Whiting, Elizabeth Donner, Rajesh RamachandranNair, Jennifer Grabowski, Nathalie Jett´e, Daniel Rodriguez Duque PII: DOI: Reference:
S0920-1211(16)30282-0 http://dx.doi.org/doi:10.1016/j.eplepsyres.2017.01.013 EPIRES 5684
To appear in:
Epilepsy Research
Received date: Revised date: Accepted date:
5-11-2016 20-12-2016 21-1-2017
Please cite this article as: Whiting, Sharon, Donner, Elizabeth, RamachandranNair, Rajesh, Grabowski, Jennifer, Jett´e, Nathalie, Duque, Daniel Rodriguez, Decreased Health Care Utilization and Health Care Costs in the Inpatient and Emergency Department Setting Following Initiation of Ketogenic Diet in Pediatric Patients: The Experience in Ontario, Canada.Epilepsy Research http://dx.doi.org/10.1016/j.eplepsyres.2017.01.013 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Decreased Health Care Utilization and Health Care Costs in the Inpatient and Emergency Department Setting Following Initiation of Ketogenic Diet in Pediatric Patients: The Experience in Ontario, Canada
*Sharon Whiting, +Elizabeth Donner, ++ Rajesh RamachandranNair, *Jennifer Grabowski , **Nathalie Jetté , o Daniel Rodriguez Duque
*Children’s Hospital of Eastern Ontario, University of Ottawa, Ottawa; +Hospital for Sick Children, Toronto; ++McMaster Children’s Hospital, Hamilton,*Children’s Hospital of Eastern Ontario, Ontario; **Department of Clinical Neurosciences, Hochkiss Brain Institute, O’Brien Institute for Public Health, University of Calgary, Calgary, oInstitute for Clinical Evaluative Sciences, Ontario;
Highlights
The ketogenic diet is an effective treatment option for epilepsy
Patients on the ketogenic diet experienced decreased health care utilization
Patients on the ketogenic diet had decreased health care costs.
Summary: Objective: To assess the change in inpatient and emergency department utilization and health care costs in children on the ketogenic diet for treatment of epilepsy. Methods: Data on children with epilepsy initiated on the ketogenic diet (KD) Jan 1, 2000 and Dec 31, 2010 at Ontario pediatric hospitals were linked to province wide inpatient, emergency department (ED) data at the Institute for Clinical Evaluative Sciences. ED and inpatient visits and costs for this cohort were compared for a maximum of 2 years (730 days) prior to diet initiation and for a maximum of 2 years (730 days) following diet initiation. KD patient were compared to matched group of children with epilepsy who did not receive the ketogenic diet (no KD). Results: Children on the KD experienced a mean decrease in ED visits of 2.5 visits per person per year [95% CI (1.5-3.4)], and a mean decrease of 0.8 inpatient visits per person per year [95% CI (0.3-1.3)], following diet initiation. They had a mean decrease in ED costs of $630 [95% CI (249-1012)] per person per year and a median decrease in inpatient costs of $1059 [IQR: 7890; p<.001] per child per year. Compared with the no KD children, children on the diet experienced a mean reduction of 2.1 ED visits per child per year [95% CI (1.0-3.2)] and a mean decrease of 0.6 [95% CI (0.1-1.1)] inpatient visits per child per year. Patients on the KD experienced a reduction of $442 [95% CI (34.4-850)] per child per year more in ED costs than the matched group. The ketogenic diet group had greater median decrease in inpatient costs per child per year than the matched group. [p<0.001]. Significance: Patients initiated on ketogenic diet, experienced decreased ED and inpatient visits as well as costs following diet initiation in Ontario, Canada. Keywords: Ketogenic Diet, Epilepsy, Health Care Utilization, Health care Costs
Introduction: Epilepsy affects approximately 15,000 children in Ontario (Ng et al 2015), Canada’s most populous province. Of these children, an estimated 30% have drug-resistant epilepsy. The high-fat, lowcarbohydrate, ketogenic diet (KD), in use since the 1920s, is an effective treatment for this patient population. (Kossoff et al 2009, Neal et al 2008). A large number of studies, (Kossoff, McGrogan 2005, Kossoff et al 2009, Sharma et al 2013) including randomized-controlled trials and a recently published meta-analysis (li 2013), indicate that the diet is effective at decreasing the number and severity of seizures. Although the diet has been adopted worldwide in the recent years and the number of countries utilizing the diet is growing (Kossoff, McGrogan 2005), there is a lack of research on the effect of KD on health care utilization patterns and costs to the health care system. The primary purpose of this study was to describe and compare ED and inpatient health care utilization patterns and associated costs of Ontario children before and after they were initiated on KD and as a secondary objective this study sought to compare utilization and costs of children who received the diet ( KD group) to utilization and costs of children with epilepsy who did not receive KD ( noKD group). We chose to focus on ED and inpatient visits as we hypothesized that use of healthcare services would decline through the use of the KD. Methods Study Population The period of interest was the decade between the years 2000 and 2010. Three tertiary care pediatric centers, The Children’s Hospital of Eastern Ontario (Ottawa), The Hospital for Sick Children (Toronto), and McMaster Children’s Hospital (Hamilton), utilized the diet during the study time frame and participated in data collection. Research Ethics Board approval was obtained for the study. A list of all patients who received KD during the study period was generated via medical records search at all three of the study locations. Search terms included epilepsy, seizure disorder, ketogenic diet, and dietician consult. This list was cross-referenced with a list kept by dieticians and a list of patients for whom an EEG had been obtained, to ensure that all patients who received KD were identified. Retrospective medical chart
data from this cohort was entered into RedCap, a secure, online, database. Only patients initiated on KD at each of the three respective centers between January 1, 2000 and December 31, 2010 were included. Given that these were the only three major tertiary pediatric centers in Ontario utilizing the KD for drugresistant epilepsy, this list of patients represents the majority of Ontario children placed on the KD during the study time frame. Specific variables of interest were: date of epilepsy diagnosis, date of diet initiation, indicator of ongoing diet at the time of database creation, and date of diet termination. In the patient charts, the exact date of diet start date was not always available; as a result, this date might have differed by a few days from the true diet start date. An encoded version of the patients’ Ontario Health Insurance Plan (OHIP) number was used to deterministically link patients to health administrative data from several Ontario-wide databases available at the Institute for Clinical Evaluative Sciences (ICES), capturing Ontario residents’ contact with the province’s universal healthcare system across time. The following databases were used: Ontario Registered Persons Database (RPDB) for demographic variable, OHIP physician claims data to identify matched controls for the children receiving the diet, Discharge Abstract Database (DAD) to obtain inpatient visits, and costs related to these visits, ERCLAIM for health care utilization in the ED setting and National Ambulatory Care Reporting System (NACRS; available starting fiscal year 2003) for the ED cost analysis. Two ED databases were used because, unlike NACRS, ERCLAIM does not contain health system costing variables but does cover the entire study period. ERCLAIM is derived from physician billing data for services covered under the OHIP that are rendered at the ED. NACRS data are captured through a reporting system that records patient visits at hospital and community based ambulatory care centers, including the ED. Costs for ED and inpatient visits were obtained using costing methods developed for health administrative data (Wodchis et al 2013). These methods capture direct healthcare costs paid out by the Ontario healthcare system. In the context of inpatient and ED visits, these methods capture the cost of a case by multiplying the case’s Resource Intensity Weight (RIW) by the average provincial cost per weighted case (CPWC). Generally speaking, the RIW is a measure of a case’s resource usage relative to the resource usage of an average case; CPWC is a measure of the expenditures generated
in treating the average case. RIW’s and CPWC’s are year and sector specific; the appropriate values were used for inpatient and ED visits. Patients in the KD database who initiated the diet between January 1, 2000 and December 31, 2010 were excluded from the study if they had an invalid OHIP number (i.e. out of province patient). Patients were also excluded if their diet status at the time of data abstraction or if date of epilepsy diagnosis were missing. Patients who died during the observation period or who were not eligible to receive OHIP coverage in the defined observation window were also excluded. If patients started the diet on multiple occasions, only the first diet initiation was used. We did not expect patients to die for reasons related to the diet and did not want to capture the healthcare use and costs related to end of life. Henceforth, the patients remaining after exclusions are referred to as the KD group. Design The index date for the KD group was defined as the date that the ketogenic diet was initiated. A look-back window and look-ahead window was defined in reference to the index date. For each patient in the KD group, the look-ahead window was defined as the minimum of two dates: date of KD termination, and two years (730 days) post index date. The length of the look-back window was mirrored, unless it extended past a patient’s date of birth. In this case, the beginning of the look-back window, the patient’s study start date, was equated to the date of birth. We compared health services utilization and cost of the KD group to the utilization and cost generated by a matched group of children with epilepsy not on the KD (noKD group). The noKD children were selected based on a 1:1 match of KD group patients. To make the two groups more comparable, matching was performed based on epilepsy diagnosis during the study period, sex, rural/non-rural status of the child’s residence at index event date (Kralj 2009), date of birth within one year, and fiscal year of study start. Additionally, each match was followed for the same amount of time that the corresponding KD group patient was followed for. Following each match for an equivalent amount of time as the corresponding case is important because the average rate at which healthcare visits are generated should not be assumed to be independent of follow-up time. Matching was performed sequentially based on the
randomly ordered patients in the KD group. Since the noKD group did not have a naturally defined index date, their index date was defined relative to the date of epilepsy diagnosis to ensure the matched pairs had epilepsy for the same amount of time in the observation window, as depicted in Figure 1. Note that amount of time between epilepsy diagnosis and Study start was not a matching factor due to limitations on the time spanned by the administrative data. Additionally, the KD group had a small number of patients who, in the data, received an epilepsy diagnosis some days after the diet start; this was likely because of the slight variation of diet start date abstracted from the charts. To identify patients diagnosed with epilepsy from the health administrative data, we used an algorithm requiring three physician claims (OHIP diagnostic code 345) at least 30 days apart in a two-year period (Tu et al 2013). This algorithm has been validated in the Ontario adult population with sensitivity 73.7 (64.8-82.5), specificity 99.8 (99.699.9), PPV 79.5 (71.1-88.0) and NPV 99.7 (99.5-99.8). As in the validation paper, the third physician billing was taken to be the epilepsy diagnosis date. Although both the KD and the noKD group were identified as having epilepsy, we could not match based on drug resistant epilepsy. Analysis The number and cost of ED and inpatient visits was obtained for the look-back window and for the lookahead window for patients in both the KD and noKD groups. Inpatient visits corresponding to a birth record were eliminated in order to maintain a fair comparison. Additionally, given that the KD is often started in an inpatient setting, inpatient admissions corresponding to the KD start were also excluded. These records were identified as the KD start records if for a given patient, the admission location was at the hospital where the diet was started, within 2 weeks of the KD start date and with a length of stay of less than seven days. The primary reason for admission in these diet start records was most often listed to be epilepsy related; dietary counselling and surveillance was most often only listed as a secondary diagnosis. It was not possible to distinguish admissions that would have occurred, regardless of diet start or not. As a result, it was unclear if diet start visits should be classified into the pre or post period. Additionally, given the limitations of the administrative data, the cost attributed to starting the diet could not be independently estimated from other co-occurring factors in the visit. Excluding the diet start visit,
ensured that we were comparing costs before and after the diet start. Inpatient visits were counted by distinct episodes of care. That is, inpatient records related to the initial cause of admission, such as transfers between hospitals or overlapping records, were not counted as a second hospitalization (but rather considered as one inpatient visit). The number of visits and cost per patient per year were calculated by dividing the appropriate utilization or cost aggregations by the length of each patient’s lookback or look-ahead window. This yielded a pre index date rate and a post index date rate. Dividing by the amount of follow-up was required as follow-up time was not uniform across patients. For each of the KD and noKD groups separately, a paired t-test (PTT) was conducted to assess whether there was a non-zero change in the mean visits/costs per person per year going from the pre to the post index event period; the Wilcoxon Signed Rank Test (WSRT) was used in the case of skewed data. All tests and confidence intervals were two sided and were reported at an alpha=0.05 level. In order to compare the average change in the KD group to the average change in the noKD group, the average difference in rates for the KD group was compared with the average difference in rates for the noKD group by making use of a two sample t-test (TSTT). The Satterthwaite correction (TSTT-S) was used in cases where the two groups had statistically different variances. The Wilcoxon Rank Sum Test (WRST) was used to compare the median change in utilization and cost between the groups in cases of skewed data. All costs were adjusted for inflation to 2013 Canadian dollar values using the healthcare specific yearly Consumer Price Index reported by Statistics Canada. All analyses were performed using SAS version 9.3 (SAS Corp, Cary, NC). Results Of the 192 records linked on patients who received KD in the study period, 8 records were excluded because it was the patient’s second diet start or because they did not have a valid OHIP number, 10 were excluded because of missing information on their diet status at the end of the study, because of missing information on their date of epilepsy diagnosis or because they were deceased during the study period. Last, 8 were excluded because they did not meet OHIP eligibility during their defined observation
window. Exclusions are grouped due to small cell sizes. Table 1 and Table 2 present demographic information for the KD group and for the noKD group. The majority of KD patients were on the diet for more than two years and most patients started the diet within six years of being diagnosed with epilepsy. From table 1, we see that the length of time between index date and epilepsy diagnosis date is longer for the KD group than for the noKD group, this is as expected given that we could only ensure that patients had epilepsy for the same amount of time in the study window, not beyond. It is also notable that using the epilepsy ascertainment algorithm, all patients in the KD group were identified correctly as having epilepsy. Before diet start, children in the KD group had a total of 841 ED visits, this changed to a total of 490 visits after the diet. They also had 366 inpatient visits before the diet and 234 after the diet. KD group patients with a study start after fiscal year 2003 (N=108) had ED costs totaling $140,848 before diet start and $92,762 after diet start. Those with study start after fiscal year 2002 (N=120), also had inpatient costs totaling $1,736,267 before diet start and $1,419,483 after diet start. Utilization Table 3 shows the mean yearly ED and inpatient visits before and after the index date, as well as the prepost change. Only the KD group patients experienced a significant decrease in mean yearly ED visits per person per year, and both KD and noKD group patients experienced a significant reduction in mean inpatient visits per person per year. Additionally, KD group patients experienced a greater reduction in mean ED and Inpatient visits per person per year than noKD group patients. This difference in reductions was 2.1 [TSST-S; (1.0-3.2)] ED visits per person per year and 0.6 [TSST- S; (0.1-1.1)] inpatient visits per person per year more for the KD compared to the noKD group. Costs Table 4 shows the mean yearly ED and median yearly inpatient costs per person before and after the index event date, as well as the pre-post change. Both KD and noKD group children experienced a significant reduction in mean ED costs per person per year. The KD group experienced an added mean reduction of $442 [TSTT-S; (35.8-848.5)] per person per year over the noKD group. Only patients in the
KD group experienced a median reduction in yearly inpatient costs [WSRT P<0.001]. By comparing the patients in the KD group with the patients in the noKD group, we found that the median decrease in yearly inpatient costs was larger in the KD group [WRST; P<0.005] than in the noKD group, who did not experience a change. Overall, we saw that KD group patients experienced a decrease in both the number of visits and in costs across the ED and inpatient setting. These reductions were greater than the reductions, or lack of reductions, observed in the noKD group patients. Discussion We can conclude that in the province of Ontario, children on KD showed an overall decreased healthcare burden following diet initiation, as measured by ED visits and hospital admissions. This decreased healthcare utilization following diet initiation also translated into a reduction in healthcare costs in both the ED and inpatient setting. To our knowledge, this is the first study to describe the healthcare utilization patterns and associated costs in Ontario in a cohort of KD patients before and after they started the KD. The matched noKD group allowed us to observe that the health care utilization (ED visits and inpatient visits) did not decrease in the post index event window for this patient population in the same way as it did in the KD group. Although the noKD group was an imperfect control group, it did help address an important question: can the decreased utilization costs in the KD group be attributed to factors beyond secular trends? By observing the health care utilization and costs of this similar group of children, the noKD group, across time, it is apparent that the change observed in the KD group is unlikely to be solely explained by the change in health care utilization that children might experience as they age. This is consistent with the idea that the diet itself, and not other factors, led to a decrease in health care utilization and costs, though causation cannot be attributed with this study design. This study has several strengths. During the study time frame, only the three centers involved in this study were offering the ketogenic diet to treat pediatric patients with drug-resistant epilepsy in Ontario; thus, the study cohort represents the majority of pediatric epilepsy patients initiated on the diet in the entire
province. The possibility of linking these patients to their ED and inpatient records across the whole province is also an important feature of this study. Also, the use of a comparator group, the noKD group, allowed us to observe if similar children not on KD experienced a similar change in utilization. This helped us explore whether the KD was associated with a decrease in health care utilization. The EPI group is a good comparator as we were able to match on several confounders such as epilepsy status, sex, date of birth, and age at study start. Additionally, having followed the two groups for approximately the same amount of time ensures that the visits and costs are comparable across the two groups. This is especially important in patient groups that have changing level of healthcare usage across time. However, we acknowledge that this study has several limitations. First, our analysis was based on retrospectively collected data. Additionally, the costing data is not available for the entire time frame of the study, thus somewhat limiting the generalizability of the costing conclusions. The costing perspective implemented is a health systems perspective, so it does not capture the full picture of the change in costs related to ED and inpatient visits. Our selection criteria for the noKD group generated a good comparator group; however, there are several ideal characteristics that this group lacks. These include all patients in the noKD group having drug resistant epilepsy, having the same comorbidities, having the same socioeconomic status, and having an epilepsy diagnosis for the same amount of time, not just within the study window. These matching criteria were not attainable for several reasons. First, limitations in the granularity of the data did not allow for the identification of certain important factors, such as drug resistant epilepsy status. Second, the smaller population size of children with epilepsy did not make it feasible to simultaneously match on a large variety of confounders. Last, limitations in the time spanned by the administrative data led to not being able to match based on length of time between epilepsy diagnosis and index date, as a result, it was only possible to guarantee that matched pairs had an epilepsy diagnosis for the same amount of time in the study window. The results of the matching yielded a noKD group who, on average, had an epilepsy diagnosis for a shorter period of time than the KD group; we might expect these patients to have higher healthcare utilization in the period following their diagnosis. As a result, the decrease in utilization experienced by the KD group could be even greater than the
decrease experienced by the noKD group, if amount of time with epilepsy had been a matching factor. Overall, the most optimal matching criteria were selected within the parameters of what was feasible. Having selected only one match per case, also limits the precision of the estimated effect differences. Lastly, the PPV of the algorithm used to identify the noKD children suggests that a portion of the noKD group does not have a clinical diagnosis of epilepsy but had only displayed epilepsy-like symptoms, though the algorithm was evaluated in adults and may in fact perform more optimally in children who are more likely to be followed by specialists than are adults with epilepsy. Although we cannot be certain that that the decreased health care utilization and decreased health care costs are directly related to the ketogenic diet, this study represents an important piece of evidence for both clinicians and policy makers. Additionally, this study demonstrates the importance of studying healthcare utilization in children pre and post KD. In the future, to continue exploring the relationship between KD and healthcare utilization, it would be worthwhile to study a larger, prospective cohort, with a longer follow-up time period in order to gather further data on the relationship between KD and health care utilization patterns. It is also worthwhile doing a cost analysis that encompasses a broader perspective. In contrast, a recent publication of a 4-month study of patients with intractable epilepsy treated with the ketogenic diet and compared with those treated as usual care failed to show a favorable economic impact. While seizure control was improved in the patients on the diet, health care costs were high initially due to hospitalization for the diet. The authors used a cost utility analysis with quality adjusted years (QALYs) as a primary outcome and cost effectiveness analysis with treatment responder as the primary outcome. The authors advised interpretation with caution as the study follow-up was short and major limitations in Quality of life measures in this group of patients (de Kinderen et al 2016). A study done on the cost effectiveness of the diet vs. VNS and care as usual using a decision analytic model with a 5-year time horizon, suggested that on average the benefits of the diet and VNS failed to outweigh the costs of the therapies. However the Quality adjusted years (QUALYs) were the primary measure of effect and a major limitation was the inability of the model to predict outcomes over a longer
time period (de Kinderen et al 2015). Both of these studies use models that are very different from ours, shorter time frames and focused on quality of life measures. Many factors have to be considered in using the ketogenic diet, cost being one of them. The International League Against Epilepsy Task Force for Dietary Therapy published the minimum requirements for the diet in resource limited regions giving clinicians practical cost effective recommendations (Kossoff et al 2015) Our study suggests that while funding of the ketogenic diet is an issue, the long term economic impact in terms of emergency department utilization and inpatient costs should not be deterrents
Acknowledgements This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). The opinions, results and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES or the MOHLTC is intended or should be inferred. N. Jette is the holder of a Canada Research Chair in Neurological Health Services Research.
This research was conducted with the support of EpLink - the Epilepsy Research Program of the Ontario Brain Institute. The Ontario Brain Institute is an independent non-profit corporation, funded partially by the Ontario government. The opinions, results and conclusions are those of the authors and no endorsement by the Ontario Brain Institute is intended or should be inferred. References References
de Kinderen, R., Postulart, D., Aldenkamp, A., Evers, S., Lamberchts, D., de Louw, A., Majoie, M., Grutters, J., 2015.Cost- effectiveness of the ketogenic diet and vagus nerve stimulation for the treatment of children with intractable epilepsy. Epilepsy Res 110,119-131 de Kinderen, R., Lambrechts, D., Wijnen, B., Postulart, D., Aldenkamp, A., Majoie, M., Evers, S., 2016. An economic evaluation of the ketogenic diet versus care as usual in children and adolescents with intractable epilepsy: An interim analysis. Epilepsia 57, 41 -50
Kossoff, E., McGrogan, J. R., 2005. Worldwide use of the ketogenic diet. Epilepsia 46, 280–289 Kossoff, E., Laux, L. C., Blackford, R., Morrison, P. F., Pyzik P. L., Hamdy, R. M.,Turner,Z., Nordli D. R., 2008. When do seizures usually improve with the ketogenic diet? Epilepsia 49, 329–333. Kossoff, E. H., Zupec-Kania, B. A., Amark, P. E., Ballaban-Gil, K. R., Bergqvist A. G., Blackford, R., Buchhalter, J.,Caraballo,R.,Cross,H.,Dahlin,M.,Donner, E., Klepper,J.,Jehle,R.,Kim, liu,C.,Nation,J.,Nordli,D.,Pfeifer,H.,Rho,J.,Strafstrom,C.,Thiele,E.,Turner,Z.,Wirrell,E.,Wh eless,J.,Veggiotti, P.,Vining,E.,and the Charlie Foundation and the Practice Committee of the Child Neurology Society,2009. Optimal clinical management of children receiving the ketogenic diet: Recommendations of the International Ketogenic Diet Study Group. Epilepsia 50, 304–317. Kossoff, E., Al-Macki, N., Cervenka, M., Kim, H., Liao, J., Megaw, K., Nathan, J., Raimann, X., Rivera, R., Wiemer-Kruel, A., Williams, E., Zupec-Kania, B., 2015. What are the minimum requirements for the ketogenic diet services in resource –limited regions? Recommendations from the International League Against Epilepsy Task Force for Dietary Therapy. Epilepsia 56, 1337-1342. Kralj B., 2009. Measuring Rurality - RIO2008_BASIC: Methodology and Results, 1–20. Li, H. F., Zou, Y., Ding, G., 2013. Therapeutic success of the ketogenic diet as a treatment option for epilepsy: A meta-analysis. Iran J Pediatr 23, 613–620. Neal, E. G., Chaffe, H., Schwartz, R. H., Lawson, M. S., Edwards, N., Fitzsimmons, G., Whitney,A.,Cross, J. H., 2008. The ketogenic diet for the treatment of childhood epilepsy: a randomised controlled trial. The Lancet Neurol7, 500–506. Ng, R., Maxwell, C. J., Yates E. ., Nylen, K., Antflick, J, Jette, N., Bronskilll, S. E., 2015. Brain Disorders in Ontario: Prevalence, Incidence and Costs from Health Administrative Data. Sharma, S., Sankhyan, N., Gulati, S., Agarwala, A., 2013. Use of the modified Atkins diet for treatment of refractory childhood epilepsy: A randomized controlled trial. Epilepsia 54, 481–486. Tu,K.,Wang, M., Jaakkimainen, R. L., Butt, D., Ivers N M, Young J.,Green,D., Jetté, N., 2014. Assessing the validity of using administrative data to identify patients with epilepsy. Epilepsia 55, 335–343. Wodchis, W., Bushmeneva, K., Nikitovic, M., McKillop, I., 2013. Guidelines on person level cost using administrative databases in Ontario, 1(May).
Figure 1: Assigning the study time frame definitions.
KD Group
noKD Group
Case1 Study Start Date before Epilepsy Diagnosis Date before
Epilepsy Diagnosis
Study Start
Diet Start Date
L1
L2
Diet Start
Study End
Epilepsy Diagnosis
Study Start
L1
L3
L2
Study End
Index Date
L3
Note: L refers to the time between two events.
Case2* Diet Start before
Study Start
Diet Start
Epilepsy Diagnosis
Study End
Study Start
Index Date
Epilepsy Diagnosis
Study End
Epilepsy Diagnosis Date L1
L2
L1
L3
L2
L3
= Study
Diet Start
Note: L refers to the time between two events.
Case3** Epilepsy Diagnosis Date before Diet Start
Epilepsy Diagnosis
L1
Study Start
L2
Diet Start
L3
Study End
Epilepsy Diagnosis
Start
L2
Study End
L3
Note: L refers to the time between two events. *A very small number of patients were in this group and the time between the two events was within days of each other. Recall, the date of diet start date came from abstracted chart data. It was not always possible to find the exact diet start date, so in some cases there is slight variability around the true start date. This led to a small number of patients being assigned a diet start that is a few days before their diagnosis date. ** For case three, matching was only performed on L2 and L3. That is, we only ensured that patients were diagnosed with epilepsy for the same amount of time inside the study window, not outside. This was done because some patients in the KD group had many years between epilepsy diagnosis and study start; the administrative data did not span the right amount of time in order for the controls to be matched based on time between epilepsy diagnosis and study start.
Table 1: Patient Characteristics *Chi-square test for homogeneity: P=0.0518 **Interval (years) between epilepsy diagnosis to start of Ketogenic diet (KD) *** A small number of patients were started on the KD before they were given a confirmed diagnosis of epilepsy. These patients are not included in this calculation.
KD Group Total N =166
N (%)
Age at Index Event Date
No KD Group Mean (SD)
Median (IQR)
5.8 (4.5)
5 (7)
N (%)
under 2 years
48(28)
50(30)
2-5 years
26(16)
23(14)
5-10 years
57(34)
58(35)
10+ years
35(21)
36(21)
Mean (SD)
Median (IQR)
5.8 (4.5)
5 (6)
1.3(0.7)***
1.3(1.4)***
Gender Male
90 (54)
90 (54)
Female
76 (46)
76 (46)
153 (92)
153 (92)
27 (16)
41 (25)
27 (16)
31 (19)
36 (22)
33 (20)
32 (19)
36 (22)
44 (26)
24 (15)
Non-Rural Income Quintile at Index Event Date* 1 (Lowest) 2 3 4 5 (Highest)
Time between Diagnosis and index event date (years)**
4.5(3.8)***
3.5 (5.1)***
< 1.5 49 (30)
90 (54)
24 (14)
76 (46)
1.5 - 3 3-6 48 (29) 6-9 22 (13) 9-12
12 +
14 (8) 9 (5)
Cause of Epilepsy Idiopathic
26 (15)
Chromosomal abnormality
16 (11)
Specific genetic syndrome
15 (9)
Other brain malformation
10 (6)
Other causes
44 (19)
Unknown
68 (40)
Epileptic Encephalopathy Yes
57 (35)
N=3 Missing ILAE classification Localization-related
33 (19)
Generalized
88 (54)
Undetermined
21 (12)
Unclassifiable
24 (14)
Table 2: Diet Characteristics for KD group N (%)
Mean (SD)
Median (IQR)
No. of AED’s Prior to KD start
5 (2.1)
5 (3)
Duration of diet (years)
1.5(0.68)*
2 (1)
< 0.5
30(18)
0.5-1
12(7)
1-1.5
15(9)
1.5-2
10(6)
2 or more
99(69)
Site of diet initiation CHEO
24(14)
McMaster
52(31)
SickKids
90(54)
Type of Diet ** Classic
126(76)
MCT
41(25)
Modified Atkins
14(8)
*Patients who were on the diet for longer than two years were calculated as being on it for two years **Several patients were on 2 types of diet at one time
Table 3: ED and Inpatient visits per person per year.
Before
After
Before
After
Group
Statistic
KD
Mean
KD
(SE)
ED Visits per Year
N
Inpatient Visits per Year
N
4.6 (0.60)
166
2.1 (0.27)
166
Median (IQR)
2.4 (5.8)
Mean
2.1 (0.25)
(SE)
1 (2.5) 166
1.3 (0.24)
166
Median (IQR)
1 (2.9)
0.5 (1.3)
KD
Mean Change*
Decrease: 2.5 CI (1.5-3.4)
Decrease: 0.8 CI (0.3-1.3)
166
noKD
Mean
1.7 (0.22)
0.5(0.09)
166
noKD
noKD
(SE)
Median (IQR)
0.5 (2.3)
Mean
1.3 (0.19)
(SE)
166
0.1 (0.5) 166
0.3 (0.06)
166
Median (IQR)
0 (1.5)
0 (0)
Mean Change*
Decrease: 0.4 (-0.1-0.9)
Decrease: 0.2 (0.05 -0.04) 166
*Decrease/Increase only refers to the observed direction of the change, not whether this change was statistically significant. **SE=standard error; IQR= interquartile range
Table 4: ED and Inpatient system costs per person per year
Group
Before KD
After
KD
KD
Statistic
ED Costs ($) per Year
N*
Median Inpatient
N**
(SE)
1354 (174)
108
18187 (3260)
120
Median (IQR)
706 (1966)
Mean
(SE)
724 (138)
Median (IQR)
321 (752)
Change
Mean Decrease:
Mean
3813 (14333) 108
After
noKD
noKD
Mean
(SE)
469 (86)
Median (IQR)
155 (593)
Mean
(SE)
281 (56)
Median (IQR)
77 (336)
Change
Mean Decrease:
120
1195 (5391) 108
630 CI (249-1012) Before noKD
19093 (6631)
Median Decrease:
120
1059 (7890) 108
5922 (2131)
120
0 (1557) 108
2108 (899)
120
0 (0) 108
Median Decrease:
120
188 CI (42-334)
0 (55)
*For ED costs, patients were excluded from the analysis if their study start date was before April 1st 2003, as ED costing variables are not available prior to this date. **For Inpatient costs, patients were excluded from the analysis if their study start date was before April 1st 2002, as inpatient costing variables are not available prior to this date.
S0920-1211(16)30282-0 http://dx.doi.org/doi:10.1016/j.eplepsyres.2017.01.013 EPIRES 5684
To appear in:
Epilepsy Research
Received date: Revised date: Accepted date:
5-11-2016 20-12-2016 21-1-2017
Please cite this article as: Whiting, Sharon, Donner, Elizabeth, RamachandranNair, Rajesh, Grabowski, Jennifer, Jett´e, Nathalie, Duque, Daniel Rodriguez, Decreased Health Care Utilization and Health Care Costs in the Inpatient and Emergency Department Setting Following Initiation of Ketogenic Diet in Pediatric Patients: The Experience in Ontario, Canada.Epilepsy Research http://dx.doi.org/10.1016/j.eplepsyres.2017.01.013 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Decreased Health Care Utilization and Health Care Costs in the Inpatient and Emergency Department Setting Following Initiation of Ketogenic Diet in Pediatric Patients: The Experience in Ontario, Canada
*Sharon Whiting, +Elizabeth Donner, ++ Rajesh RamachandranNair, *Jennifer Grabowski , **Nathalie Jetté , o Daniel Rodriguez Duque
*Children’s Hospital of Eastern Ontario, University of Ottawa, Ottawa; +Hospital for Sick Children, Toronto; ++McMaster Children’s Hospital, Hamilton,*Children’s Hospital of Eastern Ontario, Ontario; **Department of Clinical Neurosciences, Hochkiss Brain Institute, O’Brien Institute for Public Health, University of Calgary, Calgary, oInstitute for Clinical Evaluative Sciences, Ontario;
Highlights
The ketogenic diet is an effective treatment option for epilepsy
Patients on the ketogenic diet experienced decreased health care utilization
Patients on the ketogenic diet had decreased health care costs.
Summary: Objective: To assess the change in inpatient and emergency department utilization and health care costs in children on the ketogenic diet for treatment of epilepsy. Methods: Data on children with epilepsy initiated on the ketogenic diet (KD) Jan 1, 2000 and Dec 31, 2010 at Ontario pediatric hospitals were linked to province wide inpatient, emergency department (ED) data at the Institute for Clinical Evaluative Sciences. ED and inpatient visits and costs for this cohort were compared for a maximum of 2 years (730 days) prior to diet initiation and for a maximum of 2 years (730 days) following diet initiation. KD patient were compared to matched group of children with epilepsy who did not receive the ketogenic diet (no KD). Results: Children on the KD experienced a mean decrease in ED visits of 2.5 visits per person per year [95% CI (1.5-3.4)], and a mean decrease of 0.8 inpatient visits per person per year [95% CI (0.3-1.3)], following diet initiation. They had a mean decrease in ED costs of $630 [95% CI (249-1012)] per person per year and a median decrease in inpatient costs of $1059 [IQR: 7890; p<.001] per child per year. Compared with the no KD children, children on the diet experienced a mean reduction of 2.1 ED visits per child per year [95% CI (1.0-3.2)] and a mean decrease of 0.6 [95% CI (0.1-1.1)] inpatient visits per child per year. Patients on the KD experienced a reduction of $442 [95% CI (34.4-850)] per child per year more in ED costs than the matched group. The ketogenic diet group had greater median decrease in inpatient costs per child per year than the matched group. [p<0.001]. Significance: Patients initiated on ketogenic diet, experienced decreased ED and inpatient visits as well as costs following diet initiation in Ontario, Canada. Keywords: Ketogenic Diet, Epilepsy, Health Care Utilization, Health care Costs
Introduction: Epilepsy affects approximately 15,000 children in Ontario (Ng et al 2015), Canada’s most populous province. Of these children, an estimated 30% have drug-resistant epilepsy. The high-fat, lowcarbohydrate, ketogenic diet (KD), in use since the 1920s, is an effective treatment for this patient population. (Kossoff et al 2009, Neal et al 2008). A large number of studies, (Kossoff, McGrogan 2005, Kossoff et al 2009, Sharma et al 2013) including randomized-controlled trials and a recently published meta-analysis (li 2013), indicate that the diet is effective at decreasing the number and severity of seizures. Although the diet has been adopted worldwide in the recent years and the number of countries utilizing the diet is growing (Kossoff, McGrogan 2005), there is a lack of research on the effect of KD on health care utilization patterns and costs to the health care system. The primary purpose of this study was to describe and compare ED and inpatient health care utilization patterns and associated costs of Ontario children before and after they were initiated on KD and as a secondary objective this study sought to compare utilization and costs of children who received the diet ( KD group) to utilization and costs of children with epilepsy who did not receive KD ( noKD group). We chose to focus on ED and inpatient visits as we hypothesized that use of healthcare services would decline through the use of the KD. Methods Study Population The period of interest was the decade between the years 2000 and 2010. Three tertiary care pediatric centers, The Children’s Hospital of Eastern Ontario (Ottawa), The Hospital for Sick Children (Toronto), and McMaster Children’s Hospital (Hamilton), utilized the diet during the study time frame and participated in data collection. Research Ethics Board approval was obtained for the study. A list of all patients who received KD during the study period was generated via medical records search at all three of the study locations. Search terms included epilepsy, seizure disorder, ketogenic diet, and dietician consult. This list was cross-referenced with a list kept by dieticians and a list of patients for whom an EEG had been obtained, to ensure that all patients who received KD were identified. Retrospective medical chart
data from this cohort was entered into RedCap, a secure, online, database. Only patients initiated on KD at each of the three respective centers between January 1, 2000 and December 31, 2010 were included. Given that these were the only three major tertiary pediatric centers in Ontario utilizing the KD for drugresistant epilepsy, this list of patients represents the majority of Ontario children placed on the KD during the study time frame. Specific variables of interest were: date of epilepsy diagnosis, date of diet initiation, indicator of ongoing diet at the time of database creation, and date of diet termination. In the patient charts, the exact date of diet start date was not always available; as a result, this date might have differed by a few days from the true diet start date. An encoded version of the patients’ Ontario Health Insurance Plan (OHIP) number was used to deterministically link patients to health administrative data from several Ontario-wide databases available at the Institute for Clinical Evaluative Sciences (ICES), capturing Ontario residents’ contact with the province’s universal healthcare system across time. The following databases were used: Ontario Registered Persons Database (RPDB) for demographic variable, OHIP physician claims data to identify matched controls for the children receiving the diet, Discharge Abstract Database (DAD) to obtain inpatient visits, and costs related to these visits, ERCLAIM for health care utilization in the ED setting and National Ambulatory Care Reporting System (NACRS; available starting fiscal year 2003) for the ED cost analysis. Two ED databases were used because, unlike NACRS, ERCLAIM does not contain health system costing variables but does cover the entire study period. ERCLAIM is derived from physician billing data for services covered under the OHIP that are rendered at the ED. NACRS data are captured through a reporting system that records patient visits at hospital and community based ambulatory care centers, including the ED. Costs for ED and inpatient visits were obtained using costing methods developed for health administrative data (Wodchis et al 2013). These methods capture direct healthcare costs paid out by the Ontario healthcare system. In the context of inpatient and ED visits, these methods capture the cost of a case by multiplying the case’s Resource Intensity Weight (RIW) by the average provincial cost per weighted case (CPWC). Generally speaking, the RIW is a measure of a case’s resource usage relative to the resource usage of an average case; CPWC is a measure of the expenditures generated
in treating the average case. RIW’s and CPWC’s are year and sector specific; the appropriate values were used for inpatient and ED visits. Patients in the KD database who initiated the diet between January 1, 2000 and December 31, 2010 were excluded from the study if they had an invalid OHIP number (i.e. out of province patient). Patients were also excluded if their diet status at the time of data abstraction or if date of epilepsy diagnosis were missing. Patients who died during the observation period or who were not eligible to receive OHIP coverage in the defined observation window were also excluded. If patients started the diet on multiple occasions, only the first diet initiation was used. We did not expect patients to die for reasons related to the diet and did not want to capture the healthcare use and costs related to end of life. Henceforth, the patients remaining after exclusions are referred to as the KD group. Design The index date for the KD group was defined as the date that the ketogenic diet was initiated. A look-back window and look-ahead window was defined in reference to the index date. For each patient in the KD group, the look-ahead window was defined as the minimum of two dates: date of KD termination, and two years (730 days) post index date. The length of the look-back window was mirrored, unless it extended past a patient’s date of birth. In this case, the beginning of the look-back window, the patient’s study start date, was equated to the date of birth. We compared health services utilization and cost of the KD group to the utilization and cost generated by a matched group of children with epilepsy not on the KD (noKD group). The noKD children were selected based on a 1:1 match of KD group patients. To make the two groups more comparable, matching was performed based on epilepsy diagnosis during the study period, sex, rural/non-rural status of the child’s residence at index event date (Kralj 2009), date of birth within one year, and fiscal year of study start. Additionally, each match was followed for the same amount of time that the corresponding KD group patient was followed for. Following each match for an equivalent amount of time as the corresponding case is important because the average rate at which healthcare visits are generated should not be assumed to be independent of follow-up time. Matching was performed sequentially based on the
randomly ordered patients in the KD group. Since the noKD group did not have a naturally defined index date, their index date was defined relative to the date of epilepsy diagnosis to ensure the matched pairs had epilepsy for the same amount of time in the observation window, as depicted in Figure 1. Note that amount of time between epilepsy diagnosis and Study start was not a matching factor due to limitations on the time spanned by the administrative data. Additionally, the KD group had a small number of patients who, in the data, received an epilepsy diagnosis some days after the diet start; this was likely because of the slight variation of diet start date abstracted from the charts. To identify patients diagnosed with epilepsy from the health administrative data, we used an algorithm requiring three physician claims (OHIP diagnostic code 345) at least 30 days apart in a two-year period (Tu et al 2013). This algorithm has been validated in the Ontario adult population with sensitivity 73.7 (64.8-82.5), specificity 99.8 (99.699.9), PPV 79.5 (71.1-88.0) and NPV 99.7 (99.5-99.8). As in the validation paper, the third physician billing was taken to be the epilepsy diagnosis date. Although both the KD and the noKD group were identified as having epilepsy, we could not match based on drug resistant epilepsy. Analysis The number and cost of ED and inpatient visits was obtained for the look-back window and for the lookahead window for patients in both the KD and noKD groups. Inpatient visits corresponding to a birth record were eliminated in order to maintain a fair comparison. Additionally, given that the KD is often started in an inpatient setting, inpatient admissions corresponding to the KD start were also excluded. These records were identified as the KD start records if for a given patient, the admission location was at the hospital where the diet was started, within 2 weeks of the KD start date and with a length of stay of less than seven days. The primary reason for admission in these diet start records was most often listed to be epilepsy related; dietary counselling and surveillance was most often only listed as a secondary diagnosis. It was not possible to distinguish admissions that would have occurred, regardless of diet start or not. As a result, it was unclear if diet start visits should be classified into the pre or post period. Additionally, given the limitations of the administrative data, the cost attributed to starting the diet could not be independently estimated from other co-occurring factors in the visit. Excluding the diet start visit,
ensured that we were comparing costs before and after the diet start. Inpatient visits were counted by distinct episodes of care. That is, inpatient records related to the initial cause of admission, such as transfers between hospitals or overlapping records, were not counted as a second hospitalization (but rather considered as one inpatient visit). The number of visits and cost per patient per year were calculated by dividing the appropriate utilization or cost aggregations by the length of each patient’s lookback or look-ahead window. This yielded a pre index date rate and a post index date rate. Dividing by the amount of follow-up was required as follow-up time was not uniform across patients. For each of the KD and noKD groups separately, a paired t-test (PTT) was conducted to assess whether there was a non-zero change in the mean visits/costs per person per year going from the pre to the post index event period; the Wilcoxon Signed Rank Test (WSRT) was used in the case of skewed data. All tests and confidence intervals were two sided and were reported at an alpha=0.05 level. In order to compare the average change in the KD group to the average change in the noKD group, the average difference in rates for the KD group was compared with the average difference in rates for the noKD group by making use of a two sample t-test (TSTT). The Satterthwaite correction (TSTT-S) was used in cases where the two groups had statistically different variances. The Wilcoxon Rank Sum Test (WRST) was used to compare the median change in utilization and cost between the groups in cases of skewed data. All costs were adjusted for inflation to 2013 Canadian dollar values using the healthcare specific yearly Consumer Price Index reported by Statistics Canada. All analyses were performed using SAS version 9.3 (SAS Corp, Cary, NC). Results Of the 192 records linked on patients who received KD in the study period, 8 records were excluded because it was the patient’s second diet start or because they did not have a valid OHIP number, 10 were excluded because of missing information on their diet status at the end of the study, because of missing information on their date of epilepsy diagnosis or because they were deceased during the study period. Last, 8 were excluded because they did not meet OHIP eligibility during their defined observation
window. Exclusions are grouped due to small cell sizes. Table 1 and Table 2 present demographic information for the KD group and for the noKD group. The majority of KD patients were on the diet for more than two years and most patients started the diet within six years of being diagnosed with epilepsy. From table 1, we see that the length of time between index date and epilepsy diagnosis date is longer for the KD group than for the noKD group, this is as expected given that we could only ensure that patients had epilepsy for the same amount of time in the study window, not beyond. It is also notable that using the epilepsy ascertainment algorithm, all patients in the KD group were identified correctly as having epilepsy. Before diet start, children in the KD group had a total of 841 ED visits, this changed to a total of 490 visits after the diet. They also had 366 inpatient visits before the diet and 234 after the diet. KD group patients with a study start after fiscal year 2003 (N=108) had ED costs totaling $140,848 before diet start and $92,762 after diet start. Those with study start after fiscal year 2002 (N=120), also had inpatient costs totaling $1,736,267 before diet start and $1,419,483 after diet start. Utilization Table 3 shows the mean yearly ED and inpatient visits before and after the index date, as well as the prepost change. Only the KD group patients experienced a significant decrease in mean yearly ED visits per person per year, and both KD and noKD group patients experienced a significant reduction in mean inpatient visits per person per year. Additionally, KD group patients experienced a greater reduction in mean ED and Inpatient visits per person per year than noKD group patients. This difference in reductions was 2.1 [TSST-S; (1.0-3.2)] ED visits per person per year and 0.6 [TSST- S; (0.1-1.1)] inpatient visits per person per year more for the KD compared to the noKD group. Costs Table 4 shows the mean yearly ED and median yearly inpatient costs per person before and after the index event date, as well as the pre-post change. Both KD and noKD group children experienced a significant reduction in mean ED costs per person per year. The KD group experienced an added mean reduction of $442 [TSTT-S; (35.8-848.5)] per person per year over the noKD group. Only patients in the
KD group experienced a median reduction in yearly inpatient costs [WSRT P<0.001]. By comparing the patients in the KD group with the patients in the noKD group, we found that the median decrease in yearly inpatient costs was larger in the KD group [WRST; P<0.005] than in the noKD group, who did not experience a change. Overall, we saw that KD group patients experienced a decrease in both the number of visits and in costs across the ED and inpatient setting. These reductions were greater than the reductions, or lack of reductions, observed in the noKD group patients. Discussion We can conclude that in the province of Ontario, children on KD showed an overall decreased healthcare burden following diet initiation, as measured by ED visits and hospital admissions. This decreased healthcare utilization following diet initiation also translated into a reduction in healthcare costs in both the ED and inpatient setting. To our knowledge, this is the first study to describe the healthcare utilization patterns and associated costs in Ontario in a cohort of KD patients before and after they started the KD. The matched noKD group allowed us to observe that the health care utilization (ED visits and inpatient visits) did not decrease in the post index event window for this patient population in the same way as it did in the KD group. Although the noKD group was an imperfect control group, it did help address an important question: can the decreased utilization costs in the KD group be attributed to factors beyond secular trends? By observing the health care utilization and costs of this similar group of children, the noKD group, across time, it is apparent that the change observed in the KD group is unlikely to be solely explained by the change in health care utilization that children might experience as they age. This is consistent with the idea that the diet itself, and not other factors, led to a decrease in health care utilization and costs, though causation cannot be attributed with this study design. This study has several strengths. During the study time frame, only the three centers involved in this study were offering the ketogenic diet to treat pediatric patients with drug-resistant epilepsy in Ontario; thus, the study cohort represents the majority of pediatric epilepsy patients initiated on the diet in the entire
province. The possibility of linking these patients to their ED and inpatient records across the whole province is also an important feature of this study. Also, the use of a comparator group, the noKD group, allowed us to observe if similar children not on KD experienced a similar change in utilization. This helped us explore whether the KD was associated with a decrease in health care utilization. The EPI group is a good comparator as we were able to match on several confounders such as epilepsy status, sex, date of birth, and age at study start. Additionally, having followed the two groups for approximately the same amount of time ensures that the visits and costs are comparable across the two groups. This is especially important in patient groups that have changing level of healthcare usage across time. However, we acknowledge that this study has several limitations. First, our analysis was based on retrospectively collected data. Additionally, the costing data is not available for the entire time frame of the study, thus somewhat limiting the generalizability of the costing conclusions. The costing perspective implemented is a health systems perspective, so it does not capture the full picture of the change in costs related to ED and inpatient visits. Our selection criteria for the noKD group generated a good comparator group; however, there are several ideal characteristics that this group lacks. These include all patients in the noKD group having drug resistant epilepsy, having the same comorbidities, having the same socioeconomic status, and having an epilepsy diagnosis for the same amount of time, not just within the study window. These matching criteria were not attainable for several reasons. First, limitations in the granularity of the data did not allow for the identification of certain important factors, such as drug resistant epilepsy status. Second, the smaller population size of children with epilepsy did not make it feasible to simultaneously match on a large variety of confounders. Last, limitations in the time spanned by the administrative data led to not being able to match based on length of time between epilepsy diagnosis and index date, as a result, it was only possible to guarantee that matched pairs had an epilepsy diagnosis for the same amount of time in the study window. The results of the matching yielded a noKD group who, on average, had an epilepsy diagnosis for a shorter period of time than the KD group; we might expect these patients to have higher healthcare utilization in the period following their diagnosis. As a result, the decrease in utilization experienced by the KD group could be even greater than the
decrease experienced by the noKD group, if amount of time with epilepsy had been a matching factor. Overall, the most optimal matching criteria were selected within the parameters of what was feasible. Having selected only one match per case, also limits the precision of the estimated effect differences. Lastly, the PPV of the algorithm used to identify the noKD children suggests that a portion of the noKD group does not have a clinical diagnosis of epilepsy but had only displayed epilepsy-like symptoms, though the algorithm was evaluated in adults and may in fact perform more optimally in children who are more likely to be followed by specialists than are adults with epilepsy. Although we cannot be certain that that the decreased health care utilization and decreased health care costs are directly related to the ketogenic diet, this study represents an important piece of evidence for both clinicians and policy makers. Additionally, this study demonstrates the importance of studying healthcare utilization in children pre and post KD. In the future, to continue exploring the relationship between KD and healthcare utilization, it would be worthwhile to study a larger, prospective cohort, with a longer follow-up time period in order to gather further data on the relationship between KD and health care utilization patterns. It is also worthwhile doing a cost analysis that encompasses a broader perspective. In contrast, a recent publication of a 4-month study of patients with intractable epilepsy treated with the ketogenic diet and compared with those treated as usual care failed to show a favorable economic impact. While seizure control was improved in the patients on the diet, health care costs were high initially due to hospitalization for the diet. The authors used a cost utility analysis with quality adjusted years (QALYs) as a primary outcome and cost effectiveness analysis with treatment responder as the primary outcome. The authors advised interpretation with caution as the study follow-up was short and major limitations in Quality of life measures in this group of patients (de Kinderen et al 2016). A study done on the cost effectiveness of the diet vs. VNS and care as usual using a decision analytic model with a 5-year time horizon, suggested that on average the benefits of the diet and VNS failed to outweigh the costs of the therapies. However the Quality adjusted years (QUALYs) were the primary measure of effect and a major limitation was the inability of the model to predict outcomes over a longer
time period (de Kinderen et al 2015). Both of these studies use models that are very different from ours, shorter time frames and focused on quality of life measures. Many factors have to be considered in using the ketogenic diet, cost being one of them. The International League Against Epilepsy Task Force for Dietary Therapy published the minimum requirements for the diet in resource limited regions giving clinicians practical cost effective recommendations (Kossoff et al 2015) Our study suggests that while funding of the ketogenic diet is an issue, the long term economic impact in terms of emergency department utilization and inpatient costs should not be deterrents
Acknowledgements This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). The opinions, results and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES or the MOHLTC is intended or should be inferred. N. Jette is the holder of a Canada Research Chair in Neurological Health Services Research.
This research was conducted with the support of EpLink - the Epilepsy Research Program of the Ontario Brain Institute. The Ontario Brain Institute is an independent non-profit corporation, funded partially by the Ontario government. The opinions, results and conclusions are those of the authors and no endorsement by the Ontario Brain Institute is intended or should be inferred. References References
de Kinderen, R., Postulart, D., Aldenkamp, A., Evers, S., Lamberchts, D., de Louw, A., Majoie, M., Grutters, J., 2015.Cost- effectiveness of the ketogenic diet and vagus nerve stimulation for the treatment of children with intractable epilepsy. Epilepsy Res 110,119-131 de Kinderen, R., Lambrechts, D., Wijnen, B., Postulart, D., Aldenkamp, A., Majoie, M., Evers, S., 2016. An economic evaluation of the ketogenic diet versus care as usual in children and adolescents with intractable epilepsy: An interim analysis. Epilepsia 57, 41 -50
Kossoff, E., McGrogan, J. R., 2005. Worldwide use of the ketogenic diet. Epilepsia 46, 280–289 Kossoff, E., Laux, L. C., Blackford, R., Morrison, P. F., Pyzik P. L., Hamdy, R. M.,Turner,Z., Nordli D. R., 2008. When do seizures usually improve with the ketogenic diet? Epilepsia 49, 329–333. Kossoff, E. H., Zupec-Kania, B. A., Amark, P. E., Ballaban-Gil, K. R., Bergqvist A. G., Blackford, R., Buchhalter, J.,Caraballo,R.,Cross,H.,Dahlin,M.,Donner, E., Klepper,J.,Jehle,R.,Kim, liu,C.,Nation,J.,Nordli,D.,Pfeifer,H.,Rho,J.,Strafstrom,C.,Thiele,E.,Turner,Z.,Wirrell,E.,Wh eless,J.,Veggiotti, P.,Vining,E.,and the Charlie Foundation and the Practice Committee of the Child Neurology Society,2009. Optimal clinical management of children receiving the ketogenic diet: Recommendations of the International Ketogenic Diet Study Group. Epilepsia 50, 304–317. Kossoff, E., Al-Macki, N., Cervenka, M., Kim, H., Liao, J., Megaw, K., Nathan, J., Raimann, X., Rivera, R., Wiemer-Kruel, A., Williams, E., Zupec-Kania, B., 2015. What are the minimum requirements for the ketogenic diet services in resource –limited regions? Recommendations from the International League Against Epilepsy Task Force for Dietary Therapy. Epilepsia 56, 1337-1342. Kralj B., 2009. Measuring Rurality - RIO2008_BASIC: Methodology and Results, 1–20. Li, H. F., Zou, Y., Ding, G., 2013. Therapeutic success of the ketogenic diet as a treatment option for epilepsy: A meta-analysis. Iran J Pediatr 23, 613–620. Neal, E. G., Chaffe, H., Schwartz, R. H., Lawson, M. S., Edwards, N., Fitzsimmons, G., Whitney,A.,Cross, J. H., 2008. The ketogenic diet for the treatment of childhood epilepsy: a randomised controlled trial. The Lancet Neurol7, 500–506. Ng, R., Maxwell, C. J., Yates E. ., Nylen, K., Antflick, J, Jette, N., Bronskilll, S. E., 2015. Brain Disorders in Ontario: Prevalence, Incidence and Costs from Health Administrative Data. Sharma, S., Sankhyan, N., Gulati, S., Agarwala, A., 2013. Use of the modified Atkins diet for treatment of refractory childhood epilepsy: A randomized controlled trial. Epilepsia 54, 481–486. Tu,K.,Wang, M., Jaakkimainen, R. L., Butt, D., Ivers N M, Young J.,Green,D., Jetté, N., 2014. Assessing the validity of using administrative data to identify patients with epilepsy. Epilepsia 55, 335–343. Wodchis, W., Bushmeneva, K., Nikitovic, M., McKillop, I., 2013. Guidelines on person level cost using administrative databases in Ontario, 1(May).
Figure 1: Assigning the study time frame definitions.
KD Group
noKD Group
Case1 Study Start Date before Epilepsy Diagnosis Date before
Epilepsy Diagnosis
Study Start
Diet Start Date
L1
L2
Diet Start
Study End
Epilepsy Diagnosis
Study Start
L1
L3
L2
Study End
Index Date
L3
Note: L refers to the time between two events.
Case2* Diet Start before
Study Start
Diet Start
Epilepsy Diagnosis
Study End
Study Start
Index Date
Epilepsy Diagnosis
Study End
Epilepsy Diagnosis Date L1
L2
L1
L3
L2
L3
= Study
Diet Start
Note: L refers to the time between two events.
Case3** Epilepsy Diagnosis Date before Diet Start
Epilepsy Diagnosis
L1
Study Start
L2
Diet Start
L3
Study End
Epilepsy Diagnosis
Start
L2
Study End
L3
Note: L refers to the time between two events. *A very small number of patients were in this group and the time between the two events was within days of each other. Recall, the date of diet start date came from abstracted chart data. It was not always possible to find the exact diet start date, so in some cases there is slight variability around the true start date. This led to a small number of patients being assigned a diet start that is a few days before their diagnosis date. ** For case three, matching was only performed on L2 and L3. That is, we only ensured that patients were diagnosed with epilepsy for the same amount of time inside the study window, not outside. This was done because some patients in the KD group had many years between epilepsy diagnosis and study start; the administrative data did not span the right amount of time in order for the controls to be matched based on time between epilepsy diagnosis and study start.
Table 1: Patient Characteristics *Chi-square test for homogeneity: P=0.0518 **Interval (years) between epilepsy diagnosis to start of Ketogenic diet (KD) *** A small number of patients were started on the KD before they were given a confirmed diagnosis of epilepsy. These patients are not included in this calculation.
KD Group Total N =166
N (%)
Age at Index Event Date
No KD Group Mean (SD)
Median (IQR)
5.8 (4.5)
5 (7)
N (%)
under 2 years
48(28)
50(30)
2-5 years
26(16)
23(14)
5-10 years
57(34)
58(35)
10+ years
35(21)
36(21)
Mean (SD)
Median (IQR)
5.8 (4.5)
5 (6)
1.3(0.7)***
1.3(1.4)***
Gender Male
90 (54)
90 (54)
Female
76 (46)
76 (46)
153 (92)
153 (92)
27 (16)
41 (25)
27 (16)
31 (19)
36 (22)
33 (20)
32 (19)
36 (22)
44 (26)
24 (15)
Non-Rural Income Quintile at Index Event Date* 1 (Lowest) 2 3 4 5 (Highest)
Time between Diagnosis and index event date (years)**
4.5(3.8)***
3.5 (5.1)***
< 1.5 49 (30)
90 (54)
24 (14)
76 (46)
1.5 - 3 3-6 48 (29) 6-9 22 (13) 9-12
12 +
14 (8) 9 (5)
Cause of Epilepsy Idiopathic
26 (15)
Chromosomal abnormality
16 (11)
Specific genetic syndrome
15 (9)
Other brain malformation
10 (6)
Other causes
44 (19)
Unknown
68 (40)
Epileptic Encephalopathy Yes
57 (35)
N=3 Missing ILAE classification Localization-related
33 (19)
Generalized
88 (54)
Undetermined
21 (12)
Unclassifiable
24 (14)
Table 2: Diet Characteristics for KD group N (%)
Mean (SD)
Median (IQR)
No. of AED’s Prior to KD start
5 (2.1)
5 (3)
Duration of diet (years)
1.5(0.68)*
2 (1)
< 0.5
30(18)
0.5-1
12(7)
1-1.5
15(9)
1.5-2
10(6)
2 or more
99(69)
Site of diet initiation CHEO
24(14)
McMaster
52(31)
SickKids
90(54)
Type of Diet ** Classic
126(76)
MCT
41(25)
Modified Atkins
14(8)
*Patients who were on the diet for longer than two years were calculated as being on it for two years **Several patients were on 2 types of diet at one time
Table 3: ED and Inpatient visits per person per year.
Before
After
Before
After
Group
Statistic
KD
Mean
KD
(SE)
ED Visits per Year
N
Inpatient Visits per Year
N
4.6 (0.60)
166
2.1 (0.27)
166
Median (IQR)
2.4 (5.8)
Mean
2.1 (0.25)
(SE)
1 (2.5) 166
1.3 (0.24)
166
Median (IQR)
1 (2.9)
0.5 (1.3)
KD
Mean Change*
Decrease: 2.5 CI (1.5-3.4)
Decrease: 0.8 CI (0.3-1.3)
166
noKD
Mean
1.7 (0.22)
0.5(0.09)
166
noKD
noKD
(SE)
Median (IQR)
0.5 (2.3)
Mean
1.3 (0.19)
(SE)
166
0.1 (0.5) 166
0.3 (0.06)
166
Median (IQR)
0 (1.5)
0 (0)
Mean Change*
Decrease: 0.4 (-0.1-0.9)
Decrease: 0.2 (0.05 -0.04) 166
*Decrease/Increase only refers to the observed direction of the change, not whether this change was statistically significant. **SE=standard error; IQR= interquartile range
Table 4: ED and Inpatient system costs per person per year
Group
Before KD
After
KD
KD
Statistic
ED Costs ($) per Year
N*
Median Inpatient
N**
(SE)
1354 (174)
108
18187 (3260)
120
Median (IQR)
706 (1966)
Mean
(SE)
724 (138)
Median (IQR)
321 (752)
Change
Mean Decrease:
Mean
3813 (14333) 108
After
noKD
noKD
Mean
(SE)
469 (86)
Median (IQR)
155 (593)
Mean
(SE)
281 (56)
Median (IQR)
77 (336)
Change
Mean Decrease:
120
1195 (5391) 108
630 CI (249-1012) Before noKD
19093 (6631)
Median Decrease:
120
1059 (7890) 108
5922 (2131)
120
0 (1557) 108
2108 (899)
120
0 (0) 108
Median Decrease:
120
188 CI (42-334)
0 (55)
*For ED costs, patients were excluded from the analysis if their study start date was before April 1st 2003, as ED costing variables are not available prior to this date. **For Inpatient costs, patients were excluded from the analysis if their study start date was before April 1st 2002, as inpatient costing variables are not available prior to this date.