High Hospitalization Rates in Survivors of Childhood Cancer: A Longitudinal Follow-Up Study Using Medical Record Linkage

Hospitalization rates over time of childhood cancer survivors (CCS) provide insight into the burden of unfavorable health conditions on CCS and health care resources. The objective of our study was to examine trends in hospitalizations of CCS and risk factors in comparison with the general population. We performed a medical record linkage study of a cohort of 1564 ≥five-year CCS with national registers. We obtained a random sample of the general population matched on year of birth, gender and calendar year per CCS retrieved. We quantified and compared hospitalization rates of CCS and reference persons from 1995 until 2005, and we analyzed risk factors for hospitalization within the CCS cohort with multivariable Poisson models. We retrieved hospitalization information from 1382 CCS and 25583 reference persons. The overall relative hospitalization rate (RHR) was 2.2 (95%CI:1.9–2.5) for CCS compared to reference persons. CCS with central nervous system and solid tumors had highest RHRs. Hospitalization rates in CCS were increased compared to reference persons up to at least 30 years after primary diagnosis, with highest rates 5–10 and 20–30 years after primary cancer. RHRs were highest for hospitalizations due to neoplasms (10.7; 95%CI:7.1–16.3) and endocrine/nutritional/metabolic disorders (7.3; 95%CI:4.6–11.7). Female gender (P<0.001), radiotherapy to head and/or neck (P<0.001) or thorax and/or abdomen (P = 0.03) and surgery (P = 0.01) were associated with higher hospitalization rates in CCS. In conclusion, CCS have increased hospitalization rates compared to the general population, up to at least 30 years after primary cancer treatment. These findings imply a high and long-term burden of unfavorable health conditions after childhood cancer on survivors and health care resources.


Abstract
Hospitalization rates over time of childhood cancer survivors (CCS) provide insight into the burden of unfavorable health conditions on CCS and health care resources. The objective of our study was to examine trends in hospitalizations of CCS and risk factors in comparison with the general population. We performed a medical record linkage study of a cohort of 1564 five-year CCS with national registers. We obtained a random sample of the general population matched on year of birth, gender and calendar year per CCS retrieved. We quantified and compared hospitalization rates of CCS and reference persons from 1995 until 2005, and we analyzed risk factors for hospitalization within the CCS cohort with multivariable Poisson models. We retrieved hospitalization information from 1382 CCS and 25583 reference persons. The overall relative hospitalization rate (RHR) was 2.2 (95%CI:1.9-2.5) for CCS compared to reference persons. CCS with central nervous system and solid tumors had highest RHRs. Hospitalization rates in CCS were increased compared to reference persons up to at least 30 years after primary diagnosis, with highest rates 5-10 and 20-30 years after primary cancer. RHRs were highest for hospitalizations due to neoplasms (10.7; 95%CI:7.1-16.3) and endocrine/nutritional/metabolic disorders (7.3; 95%CI:4.6-11.7). Female gender (P<0.001), radiotherapy to head and/or neck (P<0.001) or thorax and/or abdomen (P = 0.03) and surgery (P = 0.01) were associated with higher hospitalization rates in CCS. In conclusion, CCS have increased hospitalization rates compared to the general population, up to at least 30 years after primary cancer treatment. These findings imply a high and long-term burden of unfavorable health conditions after childhood cancer on survivors and health care resources.

Introduction
More than 75% of children with cancer become long-term survivors [1,2]. However, childhood cancer survivors (CCS) are at increased risk of unfavorable health conditions associated with their previous cancer treatment [3][4][5][6][7][8]. Among CCS aged 45 years, the cumulative prevalence estimates for serious/disabling or life-threatening chronic health conditions is 80% [5,7,9]. It is likely that late effects of cancer treatment in steadily growing numbers of CCS do not only burden the survivors themselves but also the health care system [10].
Hospitalization rates over time provide insight into the burden of unfavorable health conditions on individuals and on health care resources [11][12][13][14][15]. Thus far, previous studies focused on hospital admissions as a measure of burden of disease in CCS but these studies only determined first hospitalization or average hospitalization rates [16][17][18][19][20][21][22][23][24][25][26][27]. Although these studies provide important insight into the long-term morbidity of CCS, no study has analyzed hospitalization rates in CCS longitudinally by taking into account all hospitalizations within one individual. Such trends are likely to be a better measure for the total burden of unfavorable health conditions [15,28]. Insight into trends and risk factors for all hospitalizations can help to further focus and prioritize long-term follow-up care for CCS at risk of unfavorable health conditions requiring hospitalizations.
Our objective was to determine trends in hospitalization rates of CCS and associated risk factors in comparison with a random reference sample from the general population. To obtain these data, we performed medical record linkage between a study cohort of CCS and Dutch administrative registers.

Study population and linkage procedure
The Emma Children's Hospital/Academic Medical Center (EKZ/AMC) childhood cancer survivor cohort is a single-center cohort study of CCS who survived at least five years since primary cancer diagnosis. Details of this study have been described previously [7,29]. Eligibility criteria were a primary childhood cancer diagnosis between 1966 and 1999 before age 18, survival at least five years since diagnosis and being alive at Jan 1 st 1995 (N = 1564). From our database, we retrieved information on patient characteristics, cancer diagnosis, all cancer treatment before the date of five-year survival, recurrences and subsequent cancers. Written informed consent was obtained from all childhood cancer patients treated in the EKZ/AMC. The Institutional Review Board of the EKZ/AMC in Amsterdam reviewed and approved the data collection for our cohort register and the study was deemed as evaluation of patient care and was therefore exempt from the need for ethical approval. In addition, after medical record linkage of the cohort to national registers the data did not include directly identifiable variables anymore and therefore data were analyzed anonymously.
Because an individual identification number is lacking in the Hospital Discharge Register (Dutch acronym: LMR), we used a two-step record linkage approach to obtain hospitalization information from CCS. We described and validated the two-step record linkage of our cohort separately [30]. Using this approach we linked 1477 of 1564 eligible CCS to the Municipal Personal Records Database (Dutch acronym: GBA), and subsequently 1382 CCS to LMR.
We additionally obtained a random reference sample of the Dutch general population (20 persons at maximum from GBA with corresponding year of birth and gender per CCS retrieved in GBA), to whom we assigned the starting date of follow-up of the corresponding CCS, i.e. five years after the date of primary cancer diagnosis of the CCS. After record linkage to LMR, the study population consisted of 1382 CCS and of 26583 matched reference persons (both 94% of individuals retrieved from GBA) [30].

Outcome definition
We assessed the total number of hospitalizations and the total time at risk in CCS and reference persons from 1995 until 2005. We retrieved the primary diagnosis of the hospitalizations coded according to International Classification of Disease version 9 -clinical modification (ICD9-CM) [31]. The LMR contains electronic information on all hospitalizations (day case admissions and clinical hospitalizations) of almost every hospital in the Netherlands from 1995 onwards (coverage >99% until 2004 and 96.7% in 2005) [32]. Uncomplicated (day case) hospitalizations for delivery are not included in the register.
Accrual of time at risk for hospitalization began at the date of five-year survival (i.e. five years after the date of primary childhood cancer diagnosis) or January 1, 1995, whichever came later. Accrual of time at risk ended at January 1, 2006. We only counted time at risk when a person was a unique individual in GBA based on gender, date of birth and postal code [30]. Time during hospitalization was also excluded from the time at risk. When multiple unique periods were available, we summed up the time of the unique periods to define the total time at risk for hospitalization.
We hypothesized that CCS having a primary cancer recurrence or ongoing primary cancer therapy beyond the date of five-year survivorship would have increased exposure to cancer treatment and would simultaneously inflate the hospitalization rate. Because our interest was in hospitalizations beyond primary cancer survival, we censored these CCS and their corresponding reference persons at the date of five-year survival or at the incidence date of first primary cancer recurrence if this happened after the date of five-year survival. Full censoring applied to 90 CCS and partial censoring applied to 35 CCS.

Statistical analysis
We determined average hospitalization rates during the complete study period of CCS compared to their matched reference persons. We calculated relative hospitalization rates (RHR) and absolute excess risks (AER) of hospitalization per 1000 person-years at risk for the total group. We did the same for CCS categories as well as for specific ICD9-CM hospitalization diagnosis groups. In subsequent analyses, we allowed for changes in hospitalization rates over time in both groups, by including follow-up time since primary cancer diagnosis or attained age as covariate in the regression model. We modeled both covariates via natural cubic splines (using 7 knots) to allow for non-linear trends over time. Parameter estimates of the individual spline components are hard to interpret. Therefore, we used graphs to describe trends. We also estimated hospitalization trends over follow-up time for CCS categories. In all analyses we used a Poisson regression model in which we corrected for recurrent hospitalizations via generalized estimating equations (GEE). We assumed an exchangeable correlation structure.
Finally, we examined risk factors associated with hospitalization rates within the cohort of CCS. In addition to follow-up time since primary cancer diagnosis, we included gender, calendar year of primary cancer diagnosis, age at primary cancer diagnosis and cancer treatments given before the date of five-year survival (cancer surgery; anthracyclines; alkylating agents; other chemotherapy; radiotherapy to the head and/or neck; radiotherapy to the thorax and/or abdomen; radiotherapy to extremities).
We again used a Poisson model, again correcting for recurrent hospitalizations via GEE. We explored effect modification between all included variables and follow-up time, between non-treatment variables (i.e. gender, calendar year of primary cancer diagnosis and age at primary cancer diagnosis) and main cancer treatment modalities (i.e. surgery, chemotherapy, radiotherapy), and between the main cancer treatments themselves. We included them in the final model when statistically significant. We modeled the effects of follow-up time, calendar year of primary cancer diagnosis and age at primary cancer diagnosis via natural cubic splines (using 4 or 5 knots) to allow for a non-linear relationship with the hospitalization rate. Besides providing P-values for the overall effects of the variables, we also present results graphically.
We performed analyses using statistical software R version (2.15.0) using survreg for average RHR and over time, and ggplot2 for the graphs. P values <0.05 were considered statistically significant.

Results
After medical record linkage, included CCS and reference persons contributed a total of 10,622 and 194,094 years of time at risk respectively ( Table 1  After applying the censoring for cancer treatment for primary cancer (recurrences) beyond five year survival, we identified 1736 hospitalizations in 1292 CCS, with an average rate of 172 hospitalizations per 1000 person-years. The hospitalization rate in matched reference persons was 79 per 1000 person-years. See S1 Methods for more details on the results of censoring. Table 2 shows average RHR and AER of CCS compared to reference persons. The overall RHR and AER were 2.2 (95%CI:1.9-2.5) and 93.3 per 1000 person-years at risk, respectively. RHRs and AERs were increased in all CCS cancer diagnosis and treatment categories, with the highest RHRs for CCS originally diagnosed with primary central nervous system (CNS) tumors (RHR:3.4;95%CI:2.7-4.4) and other solid tumors RHR:2.6;95%CI:2.0-3.5). CCS not treated with chemotherapy or radiotherapy had a RHR of 2.5 (95%CI:1.5-4.1). Radiotherapy (with or without surgery, without chemotherapy) was associated with the highest RHR (3.4;95%CI:2.2-5.0) compared to reference persons. Table 3 shows that CCS had significantly higher hospitalization rates in comparison to ref-  (Fig 1b) for CCS and reference persons. Hospitalization rates in CCS were the highest between 5-10 and 20-30 years since primary cancer diagnosis, and before the attained age of 10 and between the attained ages of 25-40 years.
We found generally similar time trends in most other CCS categories, such as in males and females, in groups based on diagnosis, in groups based on recurrence status, and in treatment groups (S2A- S2M Fig).
We identified risk factors for hospitalization within CCS, based on our final multivariate model (Table 4). In this model we included the following significant effect modifiers: follow-up time for gender (P = 0.04), radiotherapy for gender (P = 0.02), follow-up time for surgery (P = 0.06) and calendar year of primary cancer for radiotherapy (P = 0.04). Overall, hospitalization rate was not constant over follow-up time since primary cancer diagnosis (P<0.001). Hospitalization rates were higher in females (P<0.001, non-monotone trend over follow-up time, see S2N Fig

Discussion
We showed that CCS have increased hospitalization rates compared to the general population up to at least 30 years after primary cancer diagnosis. The average hospitalization rate in our CCS cohort was increased 2.2-fold compared to the general population. Especially survivors originally diagnosed with CNS and other solid tumors have an increased hospitalization rate compared to the general population. Within the CCS cohort, we identified that hospitalization rates over follow-up time were higher after treatment with surgery or radiotherapy and in females. These findings imply an increased and long-term burden of unfavorable health conditions after childhood cancer on survivors and health care resources. Since we only studied health conditions that lead to hospitalization, the true burden of unfavorable health conditions after childhood cancer is likely to be even higher. It is concerning that, after an initial decline in hospitalization rate, we observed a late enhanced increase between the 20 th and 30 th follow-up year since primary cancer diagnosis compared to the general population. Although we are limited by a hospital register that does not contain electronic data before 1995, likely explanations for this trend could be the deterioration of existing health conditions in CCS as well as new, late onset health conditions.
We found increased hospitalization rates for several hospitalization diagnosis groups including neoplasms, endocrine disorders, circulatory diseases and diseases of the eye in CCS compared to the general population. These numbers imply that several well-known adverse events in CCS translate into an increased risk of hospitalization. The increased risk of hospitalization for neoplasms will be mainly due to secondary neoplasms since we applied censoring at late primary cancer recurrences and treatment. The increased risk of secondary neoplasms in CCS is well known and primarily related to radiotherapy [33,34]. The high risk of endocrine disorders and circulatory diseases in CCS has also been described and is mainly related to local radiotherapy as well as several chemotherapeutic agents [21][22][23][35][36][37][38][39][40][41][42]. The risk of diseases of the eye is known, but less often described in CCS. Likely explanations for this increased hospitalization rate are diseases such as cataract after radiotherapy and other problems after orbital tumors, retinoblastoma and glucocorticoids, in combination with a low background risk of (hospitalization for) eye disease in the general population [43,44]. Finally, the increased rates in the diagnosis groups "factors influencing health status and contact with health services" and "symptoms, signs and abnormal findings not elsewhere specified" could be explained by clinical signs and symptoms in CCS unusual for the age range and low-threshold clinical evaluations because of anxiety for cancer recurrence. Future studies should explore the underlying diagnoses of the hospitalization diagnosis groups. The main treatment-related risk factors for hospitalization within CCS were surgery and radiotherapy, especially head and/or neck irradiation. These findings confirm previous study findings that irradiated survivors are the most vulnerable CCS risk group [45]. CCS treated with surgery only have generally been treated with extensive surgery. We confirmed previous findings within our cohort that this treatment modality is associated with an increased burden of health conditions [7]. In our analyses within the CCS cohort we found that female survivors are at increased risk of hospitalization compared to male survivors. Female gender has been previously linked to unfavorable health conditions in CCS. However, we found that males had a higher relative risk but a lower excess risk of hospitalization than females when comparing hospitalizations of CCS to the general population for both sexes [46]. Thus, the increased risk in hospitalization among females CCS in our analyses might be partly due to increased risks of hospitalization in females in the general population in this age range. Other studies that used hospital discharge registers and determined hospitalization rates in CCS in general found increased risks of hospitalization [16,19,20]. Reported risks were an odds ratio (OR) for any hospitalization of 4.4 [16], a standardized hospitalization ratio (SHR) of any hospitalization of 2.8 [19] and a hospital admission rate ratio of 1.7 per year [20]. These studies were not able to determine hospitalization rates in CCS longitudinally by taking into account all hospitalizations within one individual and did not include multivariate detailed treatmentrelated risk factor analyses. Two other studies have used questionnaires to determine average hospitalization rates in CCS [17,18]. These studies compared hospitalizations rates to available reference rates obtained from existing surveys. Both studies also found an increased risk of  (Fig 1A) and over attained age (Fig 1B). Abbreviations: CCS: childhood cancer survivors; py: person years; y: years. Hospitalization rates per 1000 person years of CCS (dotted line) and reference persons (continuous line) over follow-up time since (corresponding) date of primary childhood cancer diagnosis ( Fig 1A) and over attained age (Fig 1B). Grey areas represent 95% confidence intervals. Estimates were made with a Poisson regression model corrected for recurrent hospitalizations. See S1 Table for further information on the differential follow-up time per calendar year of primary cancer diagnosis. hospitalization in CCS (OR 1.6 and relative rate 1.9 respectively). They found an increased risk of hospitalization for CNS tumors [18] female survivors and after radiotherapy [17].
The most important strength of our study is that we determined hospitalization rates over follow-up time by taking into account repeated hospitalizations within one individual. Using this approach, we showed that hospitalization rates differ over time and show a clinically relevant late enhanced increase between the 20 th and 30 th follow-up year since primary cancer diagnosis. We were also able to link hospitalizations to complete information on an individual's previous cancer treatment. Our analysis accounted for recurrent hospitalizations, as well as for the confounding influences of age, gender, calendar period and late recurrences of primary cancer. Without applying the censoring at late primary cancer recurrences in our analyses, the average RHR would have increased substantially to 2.8, primarily due to hospitalizations in the first 5 to 10 years since cancer diagnosis (data not shown). Finally, hospitalizations in our study were prospectively registered in a national administrative register in the same way for CCS and reference persons. We therefore had an appropriate reference group, no risk of selection bias due to (non-)response and low risk of differential misclassification of the outcome.
A limitation of our study is that we could not directly link CCS to the hospital discharge register with one unique person identifier. However, by using unique time at risk based on linkage parameters, we analyzed hospitalization rates over follow-up time in a valid manner [30]. Another limitation is the differential follow-up for survivors treated in different time periods in our study. The data of CCS diagnosed in the 1990's will have contributed mostly to the 5-15 years follow-up time, while the data of CCS diagnosed in the 1970's contributed mostly to the 20 years follow-up time (S1 Table). However, when we added calendar year of primary cancer diagnosis to the model, this effect was not significant and the effect of follow-up time did not change much (S2P Fig). Because of the differential follow-up within our study design, we  could not further explore whether the effect of follow-up time differed by calendar year of diagnosis. Nevertheless, we accounted for our longitudinal design with left-truncation by including the entry-time in our statistical analyses [47]. Finally, for this study we rely on correct hospitalization registration by Dutch hospitals, although this applies to both CCS and reference persons and has been found to be acceptable by others [48] and by ourselves [30].

Conclusions
We showed that CCS have increased hospitalization rates compared to the general population up to many years after reaching adulthood, especially survivors originally diagnosed with CNS and other solid tumors. CCS treated with surgery or radiotherapy are at highest risk for hospitalization. Further refinements in the trends of hospitalization over time and evaluation of disease specific hospitalization rates in relation to cancer treatments are needed. The high and long-term burden of unfavorable health conditions on CCS and on health care resources underscores the need for awareness and knowledge about these health conditions among survivors and health care professionals.   Table. Calendar year of primary cancer diagnosis and corresponding follow-up time since primary cancer diagnosis and attained age at the end of follow-up.