Value of clinical trial narrative data to estimate the costs of adverse event management: a feasibility study

Abstract

Aims: The objective of this feasibility study was to determine the extent to which data from randomized controlled trials (RCTs) may serve as a useful source for collecting health care resource use (HCRU) for the purposes of estimating costs of managing adverse events (AEs), specifically, grade 3–4 nausea and thrombocytopenia, which may be experienced during chemotherapy treatment.

Materials and Methods: The feasibility study was conducted in four steps: (1) HCRU data were extracted from patient narratives in four phase 3 RCTs in non–small cell lung cancer; (2) missing HCRU data were imputed; (3) unit costs were applied to the resulting HCRU data set and costs of managing AEs were estimated; and (4) the overall utility of using RCT data as a source for estimating costs of AEs was evaluated.

Results: 33 nausea and 68 thrombocytopenia AEs met eligibility criteria and were evaluated in this study. Medication usage was recorded as a treatment in 76% of nausea AEs, although only 14% of the instances of medication usage included the minimum data elements required for costing. Platelet transfusions were provided in 24% of thrombocytopenia AEs; however, in only one instance were the minimum data elements recorded. Of nausea and thrombocytopenia AEs, 18% and 72%, respectively, required no missing data assumptions or imputation.

Limitations: Only two AEs were considered, and they may not be representative of all AEs in terms of suitability for use in estimating HCRU and costs of managing AEs. Not all grade 3–4 AEs met the criteria for requiring a patient narrative. HCRU data in the narratives were incomplete.

Conclusions: The usefulness of RCTs for estimating the costs of AEs may be improved by using a standardized form to collect HCRU data for key AEs, including an appropriate level of detail required to estimate costs of managing the AEs.

Keywords:

• Oncology
• cancer
• nausea
• thrombocytopenia
• adverse event management
• cost
• health care resource use

JEL CLASSIFICATION CODES:

• I10
• I00

Introduction

Cancer treatments commonly cause adverse events (AEs). Newer treatments that have a lower frequency of severe treatment-related AEs provide opportunities to improve patient quality of life and reduce costs, which is a reason AEs often are included in economic evaluations of cancer treatments. The Second Panel on Cost-Effectiveness in Health and Medicine included AEs in its impact inventory as an example of other health effects that should be considered in a cost-effectiveness analysis¹. The National Institute for Health and Care Excellence (NICE) notes that technology appraisals should identify health outcomes that are important to patients, such as health benefits and adverse effects², although NICE does not specifically mention the costs of adverse effects. Using lung cancer as an example, many economic evaluations of treatments incorporated the costs of AEs^3–9. In some economic analyses, AEs are a cost driver^8,9; in others, AEs are not a substantial component of costs^3–7. This may reflect the actual distribution of costs, but it also may reflect the methods and data sources used to estimate AE costs.

In modeled economic evaluations of new treatments, often the incidence rates of individual AEs are obtained from randomized controlled trials (RCTs). Generally, in a model, the incidence rate is treated like a probability and multiplied by the cost of managing an instance of the AE to get a treatment-specific expected cost of managing the AE. In RCTs, typically there are prespecified criteria and processes for defining, identifying, and reporting AEs; however, in modeled economic evaluations, the sources of data and approaches for estimating the cost to manage an AE vary. These include identifying the cost to manage an AE from the published literature and, if applicable, inflating and/or converting it so that it reflects the desired cost-year, currency, and health care system, sometimes without consideration of changes in practice patterns over time or differences across countries that may affect the underlying assumptions about health care resource use (HCRU). Another approach involves identifying the HCRU required to manage an instance of the AE and applying unit costs or reference costs to generate a total expected cost to manage the AE. The estimates of the HCRU data sometimes are based on assumptions, perhaps informed by one or more expert clinicians, or developed using a more formal approach such as a Delphi panel to derive consensus-based estimates of HCRU. In other models, a separate database analysis may be conducted to generate real-world estimates of the HCRU and costs to treat an AE, although these estimates may not necessarily be directly applicable to the target patient population for a new treatment, or relevant to the payer or provider evaluating the new treatment’s economic value. Although clinical trial data are not generally used for the purpose of estimating costs of AEs, there may be an opportunity to leverage the significant amount of information collected in association with AEs during RCTs to estimate the HCRU and costs of AEs.

The objective of this feasibility study was to determine the extent to which clinical trial data may serve as a useful source for collecting HCRU for the purposes of estimating costs of managing AEs (specifically, grade 3–4 nausea and thrombocytopenia) during cancer treatment. Specifically, we sought to use patient narratives in relation to applicable adverse events coded using the United States (US) Department of Health and Human Services Common Terminology Criteria for Adverse Events, Version 5.0¹⁰, to estimate such costs.

Methods

The feasibility study was conducted in four steps: (1) HCRU data were extracted from patient narratives; (2) missing HCRU data were assessed and imputed; (3) unit costs were applied to the resulting HCRU data set and costs of managing AEs were estimated; and (4) the overall utility of using RCT data as a source for estimating costs of AEs was evaluated. HCRU data were extracted from patient narratives, regardless of treatment arm, documented in a convenience set of four large phase 3 RCTs in patients with non–small cell lung cancer (NSCLC) sponsored by Eli Lilly and Company (Lilly): PARAMOUNT¹¹, REVEL¹², PRONOUNCE¹³, and PROCLAIM¹⁴. Information from the patient narratives used in this feasibility study was de-identified and coded. The data were not collected for the purposes of this feasibility study and did not allow the feasibility study personnel to ascertain the identity of the subjects for whom the patient narratives were developed. Therefore, this feasibility study was considered by the Institutional Review Board of RTI International to not constitute human subject research.

Nausea and thrombocytopenia were selected a priori for this feasibility study because they are relatively common AEs in cancer care that may be expected to require different resources for their management. HCRU data were extracted for those nausea and thrombocytopenia AEs that required submission of patient narratives, which were a subset of all nausea and thrombocytopenia AEs. Patient narratives were included in this feasibility study if at least one HCRU associated with an AE was reported.

All HCRU data, including medications, emergency department (ED) visits, inpatient hospitalizations, laboratory tests, medical procedures, and blood products, were extracted from the patient narratives. Specific rules for the extraction of HCRU variables were applied to ensure consistency between reviewers. All narratives underwent double data extraction; particularly complex narratives were reviewed by a medical oncologist to confirm appropriate interpretation of information.

To assess the suitability for costing purposes of the data extracted from patient narratives, the percentage of AEs eligible for costing that contained HCRU was reported for each AE. Essential information required for costing each HCRU was pre-specified. Essential information required for costing medications used for nausea were defined as medication strength, frequency, formulation, and prescription duration. Essential information required for costing platelet transfusions included number of units transfused and type of platelets transfused. For the purposes of costing hospitalizations, essential information required for costing purposes included an admission date and a discharge date. Essential information for the purposes of costing other resources mentioned in the narratives (e.g. ED visit) were similarly pre-specified. For medications, platelet transfusions, and inpatient hospitalizations, which were the most commonly described HCRU in the patient narratives, the extent of missing data was documented.

Once the data were extracted from the narratives and the extent of missingness was evaluated, all missing data were addressed using a series of rules, assumptions, and imputation approaches. For example, in the case of partial HCRU data (e.g. hospital admission date but no discharge date), efforts were made to obtain the missing data based on other information in the patient narrative (e.g. date of discharge assumed equal to date of death if a patient died while in the hospital) to ensure completeness of data. If medication details were not reported, desktop research was conducted to determine from the package insert the appropriate formulation, strength, frequency, and duration of use. If number of platelet units per transfusion was not recorded, the mean number of units for those AEs that did provide number of units was assumed. Detailed rules, assumptions, and data imputation methods are presented in the Supplemental Material (Exhibit 1).

Once a complete data set of HCRU was generated, unit costs were applied to generate cost estimates for each AE. Unit costs were drawn from publicly available data sources and adjusted for medical inflation to 2017 US dollars (USD) using the consumer price index¹⁵. Specifically, medications were costed using data from RedBook¹⁶, inpatient hospitalizations from Health Care Cost and Utilization Project¹⁷; ED visits from a special report from the Centers for Disease Control and Prevention¹⁸; and outpatient visits and procedures from the Centers for Medicare and Medicaid Services, including the Physician Fee Schedule¹⁹, Medicare Fee-for-Service Hospital Outpatient Prospective Payment System Ambulatory Payment Codes²⁰, and the Clinical Laboratory Fee Schedule²¹ (Supplementary Table S1). Costs were estimated for each AE for each patient, and AE costs were summarized using descriptive statistics for each AE in total and by grade. Finally, to form the basis of an evaluation of the effect of imputation on the costs of the AEs, a comparison was made of the costs of hospitalizations with complete data required for costing available from the patient narratives and costs of hospitalizations estimated using the dataset including imputed HCRU.

Results

HCRU data from patient narratives and extent of missingness

Of all grade 3–4 AEs in all four RCTs, 34% met inclusion criteria for this feasibility study (Figure 1). Of the eligible AEs, 14% were excluded because the patient narrative failed to mention any HCRU to manage the AE. A total of 101 events were included in this study (n = 33 for nausea and n = 68 for thrombocytopenia). All nausea AEs were recorded as grade 3. Of thrombocytopenia AEs, 42 were grade 3; 26 were grade 4.

Figure 1. Flow diagram for number of grade 3–4 adverse events costed, by adverse event. HCRU: health care resource use.

Hospitalizations were recorded among 79% of grade 3 nausea AEs (Table 1) and 19% of grade 3–4 thrombocytopenia AEs (Table 2). Medication usage was recorded as a treatment in 76% of the grade 3 nausea AEs, although in those instances only 14% of narratives included the minimum data elements (medication formulation and prescription duration) required for medication costing (Table 3). Although the medication start date was nearly always provided, the end date was missing in 21% of instances (Supplementary Table S2). Platelet count assessments were recorded in 97% of narratives for grade 3–4 thrombocytopenia (Table 2). Platelet transfusions were provided in 24% of AEs; however, in only one instance were the minimum data elements (number of units and type of platelets) required for transfusion costing recorded (Table 4). Of the nausea AEs, 6 (18%) required no missing data assumptions or imputation. None of the six included were narratives where medication use was recorded. Of the thrombocytopenia AEs, 49 (72%) required no imputation for missing data. Four included HCRU other than only laboratory assessments for platelet count.

Table 1. Percentages of AEs eligible for costing with information recorded in patient narratives pertaining to key HCRU: nausea.

	n (%)
Number of AEs eligible for costing	33
Number of AEs in which narratives indicated that patients received medication (a)	25 (75.8%)
Number of AEs for which narratives indicated that patients were hospitalized (b)	26 (78.8%)
Number of AEs for which narratives indicated that patients received medication (a) and were hospitalized (b)	18 (54.5%)

The 33 nausea AEs that were eligible for costing occurred in 31 patients. Key HCRU was defined as medications and inpatient hospitalizations. 100% of AEs eligible for costing recorded use of medication or hospitalization. No other resources were recorded other than seven ED visits, all of which coincided with a hospital admission, and one instance of a computed tomography scan.

Abbreviations. AE, adverse event; ED, emergency department; HCRU, health care resource use.

Table 2. Percentages of AEs eligible for costing with information recorded in patient narratives pertaining to key HCRU: thrombocytopenia.

	n (%)
Number of AEs eligible for costing	68
Number of AEs in which narratives indicated that platelet count was assessed	66 (97.1%)
Number of AEs in which narratives indicated that patients received a platelet transfusion (a)	16 (23.5%)
Number of AEs for which narratives indicated that patients were hospitalized (b)	13 (19.1%)
Number of AEs for which narratives indicated that patients received a platelet transfusion (a) and were hospitalized (b)	8 (11.8%)

The 68 thrombocytopenia AEs that were eligible for costing occurred in 53 patients. Key HCRU was defined as platelet count assessments, platelet transfusions, and inpatient hospitalizations. 100% of AEs eligible for costing recorded use of a key HCRU. No other resources were recorded other than one use of a nasal tampon, and three instances of medication use (eltrombopag, methylprednisolone, and dexamethasone).

Abbreviations. AE, adverse event; HCRU, health care resource use.

Table 3. Extent of known missing data for medication use: nausea.

	n (%)
Number of unique instances of medication use recorded in narratives	77
Number of instances for which medication formulation was not recorded	63 (81.8%)
Number of instances for which prescription duration was not recorded	33 (42.9%)
Number of instances for which formulation or duration was not recorded	66 (85.7%)

Table 4. Extent of known missing data for platelet transfusions: thrombocytopenia.

	n (%)
Number of unique instances of platelet transfusion recorded in narratives	27
Number of instances for which the number of units transfused was provided (a)	7 (25.9%)
Number of instances for which the type of platelets was specified (b)	8 (29.6%)
Number of instances for which number of units transfused (a) and type of platelets (b) were provided	1 (3.7%)

In all instances for which the medication end date was provided, the start date also was provided. Types of platelets were indicated as single donor (6), thrombocyte concentrate (1) and fresh, frozen plasma (1).

The average number of medications prescribed for a grade 3 nausea AE was 2.3 (77 medications for 33 AEs). Of the 77 records of medications prescribed for nausea, only one record indicated strength, and only one indicated frequency. Of the 25 unique medications, 16 (64%) are available, per the RedBook¹⁶, in multiple formulations (e.g. oral, intravenous, patch). Duration of the prescription (noted by start and end date) for 57% (44 of 77) of instances was available (Table 3). The percentage of records of medications where formulation and duration were both available (i.e. a semi-complete set of data to facilitate costing) was 14% (11 of 77).

The most commonly recorded HCRU for grade 3–4 thrombocytopenia was a laboratory assessment for platelet count (208 assessments for 68 AEs, average 3.06 per AE). Of the 16 grade 3-4 thrombocytopenia AEs that required a platelet transfusion, 27 transfusions were provided (average of 1.69 transfusions per AE requiring a transfusion).

Three patients received medication for thrombocytopenia. Of those three patients, medication start date and end date were provided for two patients, and only medication end date was provided for the other patient. For medications prescribed for thrombocytopenia, no other information was provided (e.g. strength, frequency).

The average number of hospitalizations incurred for a grade 3 nausea AE was 0.8 (27 hospitalizations for 33 AEs). For grade 3–4 thrombocytopenia, the average number of hospitalizations was 0.2 (13 hospitalizations for 68 AEs). Of the 27 unique hospitalizations for grade 3 nausea, 22 (81.5%) included both admission and discharge dates, which facilitated the calculation of length of stay (LOS) without requiring the use of imputed data (Table 5). For the 18 unique hospitalizations for grade 3–4 thrombocytopenia, 8 (61.5%) included both admission and discharge dates.

Table 5. Extent of missing data for duration of hospitalization.

	Nausea n (%)	Thrombocytopenia n (%)
Number of unique instances of inpatient hospitalizations recorded in narratives	27	13
Number of hospitalizations where both admission date and discharge date were provided	22 (81.5%)	8 (61.5%)
Number of hospitalizations where only the admission date was provided	4 (14.8%)	3 (23.1%)
Number of hospitalizations where only the discharge date was provided	1 (3.7%)	1 (7.7%)
Number of hospitalizations where neither the admission date nor the discharge date was provided	0 (0.0%)	1 (7.7%)

Of the 22 hospitalizations for nausea where both admission and discharge dates were provided, four patients died prior to discharge or resolution of the adverse event, in which case date of discharge was set to date of death.

Costs of nausea and thrombocytopenia from this feasibility study

Based on descriptive analyses of the complete HCRU data set, with unit costs applied to counts of each HCRU, the mean cost to treat an event of grade 3 nausea was $1,364.36 (standard deviation [SD] = $2,615.50) (Table 6), excluding hospitalization costs, and $13,684.11 (SD = $12,745.16) including hospitalization costs. The mean costs to treat an event of grade 3 thrombocytopenia, excluding and including hospitalization costs, were $78.40 (SD = $227.90) (Table 7) and $3,192.22 (SD = $10,003.50), respectively. For grade 4 thrombocytopenia events, excluding and including hospitalization costs, costs were $1,480.33 (SD = $2,071.41) and $5,690.64 (SD = $8,636.67). Thrombocytopenia grade 4 events were on average $1,402 more costly than grade 3 events excluding hospitalizations and $2,498 more costly than grade 3 events including hospitalization costs. The mean costs to treat an event of grade 3–4 thrombocytopenia combined, excluding and including hospitalization costs, were $614.44 (SD = $1,450.47) (Table 7) and $4,147.50 (SD = $9,516.61), respectively.

Table 6. Costs to treat nausea (all events recorded as Grade 3), per adverse event (2017 USD), by resource use component.

Resource use cost component (no. of AEs = 33)	Mean	SD	Median	Minimum	Maximum
Medication cost per event (including 8 events with no medication cost)	$721.13	$2,214.81	$151.73	$0	$12,731.22
Medication administration cost per event (including 21 with no medication administration cost)	$354.14	$1,064.03	$0	$0	$6,08;8.26
ED visit cost per event (including 26 with no ED visit cost)	$281.43	$550.79	$0	$0	$1,326.74
CT scan cost per event (including 32 with no CT scan cost)	$7.66	$44.00	$0	$0	$252.74
Total cost per event excluding hospitalization costs	$1,364.36	$2,615.50	$731.87	$0	$14,310.70
Hospitalization cost per event (including 7 with no hospitalization cost)	$12,319.75	$11,948.99	$8,691.13	$0	$53,112.47
Total cost per event including hospitalization costs (including 6 with no hospitalization cost)	$13,684.11	$12,745.16	$9,656.81	$28.60	$53,684.08

Results in this table are based on data from 33 events experienced by 31 patients. All results are based on the data set that includes imputed data.

Abbreviations. AE, adverse event; ED, emergency department; CT, computed tomography; SD, standard deviation; USD, US dollars.

Table 7. Costs to treat grade 3–4 thrombocytopenia per event (2017 USD), by resource use component.

Resource Use Component	Grade 3 (no. of events = 42)		Grade 4 (no. of events = 26)		All Grade 3-4 Events (no. of events = 68)
	Mean	SD	Mean	SD	Mean	SD	Median	Min	Max
Medication cost per event (including 65 with no medication cost)	$2.91	$18.84	$3.27	$15.55	$3.05	$17.53	$0	$0	$122.08
Platelet transfusion cost per event (including 52 with no platelet transfusion cost)	$49.30	$223.14	$1,393.62	$2,027.47	$563.30	$1,413.31	$0	$0	$8,282.08
Bleeding procedure cost per event (including 67 with no bleeding procedure cost)	$0	$0	$6.50	$33.13	$2.48	$20.49	$0	$0	$168.95
Laboratory assessment cost per event (including 2 with no laboratory assessment cost)	$26.20	$23.03	$25.92	$19.97	$26.09	$21.75	$17.06	$0	$102.36
ED visit cost per event (including 67 with no ED visit cost)	$0	$0	$51.03	$260.19	$19.51	$160.89	$0	$0	$1,326.74
Total cost per event excluding hospitalization costs	$78.40	$227.90	$1,480.33	$2,071.41	$614.44	$1,450.47	$25.59	$0	$8,324.73
Hospitalization cost per event (including 55 with no hospitalization cost)	$3,113.82	$9,982.51	$4,210.31	$7,442.03	$3,533.06	9,051.73	$0	$0	$53,280.88
Total cost per event including hospitalization costs (including 55 with no hospitalization cost)	$3,192.22	$10,003.50	$5,690.64	$8,636.67	$4,147.50	$9,516.61	$34.12	$8.53	$53,280.88

Results in this table are based on data from 68 events experienced by 51 patients. All results are based on the data set that includes imputed data.

Abbreviations. ED, emergency department; SD, standard deviation; USD, US dollars.

Hospitalization HCRU and cost estimates before and after addressing missing data

The LOS per hospitalization for grade 3 nausea including imputed data (n = 27) was 6.2 days (range, 2–22); excluding imputed data (n = 22), the average LOS per hospitalization was 6.9 days (range, 2–22). The average LOS per hospitalization for grade 3–4 thrombocytopenia including imputed data (n = 13) was 7.6 days (range, 3–22); excluding imputed data (n = 8), the average LOS per hospitalization was 7.4 days (range, 3–22). For nausea, the expected cost of hospitalization including imputed data was $15,057 and excluding imputed data was $16,680 (Table 8). Conversely, for thrombocytopenia, the cost estimate including imputed data was higher ($18,481) than the cost estimate excluding imputed data ($17,861).

Table 8. Impact of imputing missing data for hospitalization.

	Nausea	Thrombocytopenia
Average cost per hospitalization when including imputed values for missing data	$15,057.48 (n = 27)	$18,480.64 (n = 13)
Average cost per hospitalization when including only non-missing data	$16,679.95 (n = 22)	$17,861.20 (n = 8)

Discussion

Of the publications that included costs of AEs mentioned previously^3–9, only Hess et al.⁸ provided costs for managing nausea and thrombocytopenia, which were used in a modeled cost-effectiveness analysis. Hess et al.⁸ assumed that grade 3–4 AEs required hospitalization and assumed a cost per nausea event of $6,225 and for thrombocytopenia of $18,354 (inflated to 2017 US dollars¹⁵) based on hospitalization cost data from HCUP.net¹⁷.

In this study using patient narratives, based on descriptive analyses of the complete HCRU data set, with unit costs applied to counts of each HCRU, the mean cost to treat an event of grade 3 nausea was $13,684 and the mean cost to treat an event of grade 3–4 thrombocytopenia was $4,148 when including hospitalization costs. It is important to note that the set of AEs considered in this feasibility study does not reflect the full set of all AEs, being limited to the subset of those with narratives, which reflect the more severe grade 3–4 AEs. These results should not be assumed to be cost estimates for all AEs. Our cost for a hospitalization for nausea is higher and for thrombocytopenia was similar to the estimate provided in Hess et al.⁸; however, there was documentation of hospitalization in only 79% of patients who experienced nausea and 19% of patients who experienced thrombocytopenia in our study.

Patient narratives from RCTs provide a data source potentially useful as a basis for estimating HCRU and costs of treating AEs; however, substantial limitations exist with the use of RCT data as a source for HCRU required to manage AEs. A limitation of using RCT-based HCRU to estimate costs, generally, is that RCT-based costs are often protocol-driven. The extent to which RCT protocols influence the management of AEs experienced in clinical trials is unknown. In addition, although predefined close follow-up of AEs in an RCT in principle could be useful to estimate the costs of AEs, RCT protocols rarely prespecify collection of adequate information to estimate such costs. Furthermore, similar to HCRU data collected in phase 3 RCTs, attributing drug or other resource use to a particular AE from narratives is complicated and cannot always be done with certainty, making it difficult to develop cost estimates for specific AEs.

There are several limitations in this feasibility study. First, only two AEs were considered, and they may not be representative of all AEs in terms of the suitability of information recorded in patient narratives for use in estimating HCRU and associated costs of managing AEs. Additionally, not all grade 3–4 AEs met the criteria for requiring a patient narrative; therefore, not all AEs could be considered when assessing AE HCRU and costs. According to guidelines developed by the International Council on Harmonisation (ICH), a brief patient narrative should be developed describing each death, other serious adverse event, and other significant adverse event that is judged to be of special interest because of clinical importance. Events that were clearly unrelated to the test drug/investigational product may be omitted or described very briefly. In general, the narrative should describe the following: the nature and intensity of event; the clinical course leading up to the event, with an indication of timing relevant to test drug/investigational product administration; relevant laboratory measurements; whether the drug was stopped, and when; countermeasures; post-mortem findings; investigator’s opinion on causality and sponsor’s opinion on causality²².

Second, for those AEs for which a narrative was developed, perhaps because estimating HCRU and costs generally are not clinical trial objectives, HCRU data in the narratives were incomplete or not well documented; although, we found that the inclusion of imputed data made little difference to the results. Third, there were a few records (five for thrombocytopenia and eight for nausea) with no HCRU recorded. As it seems unlikely that AEs would require no HCRU, these AEs were excluded from the analysis and the decision was made to include in the analysis only patient narratives if at least one HCRU associated with an AE was reported. Additionally, an extensive amount of missing data was observed for medications used to treat nausea. For example, dosing frequencies (e.g. twice daily) were not recorded in the patient narratives (other than for one record, which indicated dosing should occur as needed). This may be because dose frequency might be regarded as unnecessary detail by someone writing the narrative given that many medications are prescribed on a particular schedule routinely. The same rationale would seem to be less compelling in explaining why 43% of records did not indicate the duration of the prescription for the medication, which is essential for costing purposes. Finally, our cost estimates assumed a cost per day based on the AE as the primary discharge diagnosis but there could have been additional reasons why patients were hospitalized. A larger study may be required to look at effects on hospital costs from additional diagnoses. Similarly, although the patient narratives pertained to AEs, an assumption was made that any HCRU recorded in the narrative should be attributed to the AE, when that may not have been the case. For example, in some instances, a medication not approved for treatment of nausea was recorded in the narrative for a nausea AE. Although each of these instances was reviewed by a clinical advisor, a decision was taken not to exclude any medication recorded in the narrative for an AE.

In international, multisite RCTs, there may be substantial differences in local strategies for treating AEs, practices for deciding when to admit a patient to the hospital, and availability of medications, any of which limit the generalizability of HCRU across regions. These same differences, as well as different cost approaches, such as costing inpatient hospitalization length of stay, can contribute to differences in costs across markets beyond differences in HCRU used for similar AEs. Furthermore, questions of generalizability of clinical trial data to a real-world population are applicable to HCRU data recorded in a clinical trial.

Systematically collected AE-related HCRU data in a structured format, such as a HCRU questionnaire for key AEs, may be a more fruitful approach than relying on HCRU data that may be recorded in patient narratives. Additionally, the questionnaire would need to address attribution of HCRU (e.g. whether the AE led to or intensified an inpatient hospitalization). In particular, the study questionnaire would need to note when AEs were co-occurring, recognizing that use of an HCRU (e.g. an inpatient hospitalization) may have been attributed to multiple AEs. If a standard reporting form, including clear start and end dates for use of HCRU for a particular AE, were incorporated into clinical trial data collection forms, it may provide the research team with a more complete dataset.

Conclusion

The usefulness of RCTs for collecting HCRU data to better understand the costs of AEs may be improved by using a standardized form to collect HCRU data for key AEs, including an appropriate level of detail required to estimate costs of managing the AEs (e.g. start and end date of drugs used to treat the AE, inpatient hospitalization LOS, and number of follow-up visits required until AE resolution).

Transparency

Declaration of funding

RTI Health Solutions received funding under a research contract with Eli Lilly and Company to conduct this study.

Declaration of financial/other relationships

D.M., C.L.B., and A.N. have disclosed that they are employees of RTI International, with whom Eli Lilly and Company contracted to complete this research. L.H., Y.-J.H., Z.L.C., and L.B. have disclosed that they are employees of Eli Lilly and Company. J.M.E. peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Author contributions

All authors have made substantial contributions to the study design, analysis, and interpretation of the results of this feasibility study and have contributed to and are wholly responsible for the content presented herein.

Acknowledgements

The authors would like to thank Rebecca Conroy, Matthew Lyall, Mackenzie Neighbors, and Weyinmi Nuabor for their aid in data extraction and James A. Kaye, MD, DrPH, for his guidance in interpreting complex patient narratives.