DRAFT – FOR FURTHER DISCUSSION

NOT FOR DISTRIBUTION OR CITATION

The Fair Price of Analogue Insulin when Compared

against Human Insulin: A Systematic Review of
Published Cost Effectiveness Studies and a
Mathematic Derivation of a Value-based Formula
Jing Luo, MD Instructor in Medicine Program on Regulation, Therapeutics, and Law (PORTAL)
Harvard Medical School, Brigham and Women’s Hospital

Draft Prepared for Multi-Stakeholder Meeting September 2017

Acknowledgements
I’d like to thank Mehdi Najafzadeh (Division of Pharmacoepidemiology, Brigham and
Women’s Hospital), David Beran (University of Geneva), Kasia Lipska (Yale Endocrinology)
and all the members of the Coalition for Affordable Insulin for help in building the
intellectual foundation that resulted in this Summary Report.
Acronyms
GDPpc Gross Domestic Product per
capita
LMICs Low and Middle-Income
Countries
US United States of America
WHO World Health Organization
CEA Cost effectiveness analysis
QALYs Quality-adjusted life years
ICER Incremental cost
effectiveness ratio
HbA1c Glycated hemoglobin
Executive Summary
Objective
To conduct a systematic literature review of published cost-effectiveness studies comparing
analogue versus human insulin for the treatment of diabetes. To determine a price for analogue
insulin that reflects its marginal benefits over human insulin.

Methodology
A systematic literature review of published cost-effectiveness studies comparing analogue versus
human insulin in type 1 and type 2 diabetes was conducted using the Tufts cost-effectiveness
analysis (CEA) registry. A value-based price for analogue insulin was mathematically derived.
Data extracted from the systematic review was used to provide a few concrete examples
expressing the additional price of analogue insulin products that reflect their marginal benefits
over comparable human insulin products, from the perspective of two low and middle-income
countries.

Results
Twenty out of 24 (83 percent) cost-effectiveness studies were conducted in North America or
Europe. Industry sponsored 22/24 studies (92 percent). All studies used computer simulation
models to calculate incremental cost effectiveness ratios. The Center for Outcomes Research
Diabetes (CORE) diabetes model was used in 16 (70 percent) studies. Treatment effect
assumptions varied widely between studies. The quality of evidence supporting most treatment
effect assumptions was low to moderate.
Costs of analogue insulins ranged from no difference relative to human insulins to a maximum
of 1140 percent more expensive. Twenty-one (88 percent) CEA studies found that analogue
insulins were cost-effective when compared to human insulin. The determination of cost-
effectiveness for an insulin product is related to both the funding source and the comparative
price of the analogue insulin product.
The price for analogue insulin that reflects its marginal benefits over human insulin (value-
based price) depends on 1) the expected additional clinical benefit (in quality-adjusted life years
gained), 2) the willingness to pay of a given country or health system, 3) and the expected cost
savings with respects to a reduced rate of diabetes related complications.

Recommendations
1) Policymakers in low and middle-income countries should choose long-acting
analogues only when their price(s) are comparable or slightly higher priced than
human insulins (i.e. NPH and 70/30) because available data from published modeling
studies suggest that the cost-effectiveness of an analogue insulin is sensitive to its
procured price. Long-acting analogue insulins are highly unlikely to be cost effective in
low- and middle-income countries when their price exceeds the price of human insulin.

2) Policymakers in low and middle-income countries should choose rapid-acting

analogues in type 2 diabetes only when the price difference between the analogue
and human insulin is negligible.
 1. Introduction
Insulin is a vital medicine for the millions of people worldwide who live with type 1 diabetes and
a portion of the hundreds of millions of people who live with type 2 diabetes. Despite the fact
that it has been almost 100 years since this medicine was first used clinically, the availability and
accessibility of therapeutically effective insulin products is still far from universal (1).

In wealthy countries the higher priced analogue insulin products have surpassed human insulin
products produced using recombinant DNA technology (2,3). While both types of insulin are far
safer and more effective than the products originally developed by the co-inventors of insulin, it
is debatable whether analogue insulins are in fact superior to human insulin for the majority of
people who need them (4). Evidence of clinical comparative superiority is easier to demonstrate
in the setting of type 1 diabetes, where people are more prone to developing low blood sugar, or
hypoglycemia (5). Untreated severe hypoglycemia may result in coma or, in rare cases, death. In
people with type 2 diabetes, the evidence for clinical benefit is more equivocal; however many
diabetologists and clinical experts suggest that the marginal value for these products may not
reflect their differential price (6). In some cases, analogue insulin products are five to 10 times
more expensive than comparable human insulins.

For the 20th version of the World Health Organization’s (WHO) Model List of Essential
Medicines, an application was submitted to list long-acting analogue insulins on the basis of
evidence suggesting that these medicines were both superior and cost-effective for the treatment
of type 1 diabetes. In essence, the authors were arguing that analogue insulins meet the criteria
for an essential medicine: namely one that satisfies the priority health needs of less wealthy
countries. Furthermore, the authors argued that prices for analogue insulins would soon fall, as
follow-on biosimilar insulin products enter the market in response to increased global demand.
However, the evaluating committee at the WHO felt that the benefits of analogue insulin in
terms of reduced HbA1c and episodes of hypoglycemia were only modest and did not justify the
current large differences in price.

This decision is significant because a high-level WHO expert committee did not believe that
long-acting insulin analogues are sufficiently superior when compared to human insulins to
warrant their consideration as a priority need. Additionally, their decision suggest that analogue
insulins are likely priced much higher in today’s international markets than their overall clinical
value.

What is a fair price for analogue insulin? At what price would a decision maker trying to procure
insulin for a large national health programme in a low- and middle- income country (LMIC) be
indifferent to choosing between analogue versus human insulin? This summary report attempts
to address these problems through an empirical and mathematical approach.

2. Methods
This report sought to determine the relative value of analogue insulin versus human insulin
products through two parts. The first part of this report was a systematic review of published
cost-effectiveness analysis (CEA) studies comparing analogue versus human insulin products.
The second part uses data elements extracted from the systematic review to express the value-
based price of analogue insulin relative to human insulin.
 2.1 Systematic Review of CEA Studies Comparing
Analogue versus Human Insulin

2.1.1 Data Source(s)

The Tufts CEA registry is an online searchable database of over 5,500 published cost-utility
analyses. The registry is maintained and updated by the Center of Evaluation of Value and Risk
in Health (CEVR), based at Tufts Medical Center, in Boston, United States (US). Three times a
year, the CEA registry team searches Medline using the terms, “QALYs,” “Quality,” or “Cost-
utility analysis.” Published articles in English that contain an original cost-utility estimate
measuring health benefits as quality-adjusted life years (QALYs) are included in the registry.
Reviews, methodologic pieces, and editorials are excluded from the registry.

2.1.2 Search algorithm

A search was conducted in the Tufts cost-effectiveness analysis (CEA) registry using the word,
“insulin” on 21 March 2017. Of the original 106 resulting abstracts, 81 were excluded based on
title only, because these studies did not directly compare at least one analogue insulin product
against at least one human insulin product.

For the purposes of this report, analogue insulins are newer insulin products that have been
modified to change either the onset or duration of action. Common analogue insulins include
the long-acting analogues glargine and detemir as well as the rapid-acting insulins aspart and
lispro. Human insulin products are those whose amino-acid and polypeptide composition is
indistinguishable from human insulin and is produced using recombinant DNA technology.
Common human insulin products are regular human insulin (soluble), NPH, and pre-mixed
insulin (70/30).

Of the 25 studies that underwent full-text review, two were excluded. One study compared the
cost-effectiveness of two pre-mixed insulin products, neither of which included human insulin.
The second excluded study was at this stage because it was of poor methodological quality, using
a non-standard model and six scenarios as its base case analysis.

This resulted in 23 studies that underwent full-text review and data extraction. Each study is
summarised and described in the Results section of this report.

An additional search for CEA studies was conducted on 8 August 2017 in the NHS Economic Evaluation
Database through the York Centre for Reviews and Dissemniation.

(CRD) Database. Of the 128 study titles that contained “insulin,”19 were duplicate studies (i.e.
already included in the study based on the Tufts CEA registry search). Four new qualifying
studies were found and analysis including these studies will be included in the final draft of this
report.

Finally, an additional published CEA study was identified by one of the peer reviewers of an
earlier draft of this report and included with the original 23 to yield a total of 24 studies.

2.1.3 Data Extraction

Common data elements were extracted by a single reviewer (JL) from each of the 23 cost-effectiveness
studies described above. The definition of each element or how each value was obtained is explained in
the table below (Table 1). Information was recorded using Microsoft Excel, where each CEA study
represented a single observation (row), and each data element represented a single variable (column).

Table 1.Data elements extracted from CEA

studies in this systematic review

Data element Definition or explanation

Title Self-explanatory
Country or geographic scope Self-explanatory
Lead author Self-explanatory
Journal Self-explanatory
Year Year of article publication
Sponsor Private company (e.g. Novo Nordisk, Eli Lilly
and Company, Sanofi) or governmental
agency sponsoring the study
Study Drug Generic name of study drug (usually an
analogue insulin or pre-mixed product
containing an analogue insulin)
Comparator drug (or therapy) Generic name of human insulin product
being compared
Patient population Type 1 or type 2 diabetes mellitus
Type of model Name of computer simulation model used to
calculate incremental costs, incremental
QALYs and cost-effectiveness ratios. For
example, IMS’ CORE Diabetes model
(explained in further detail below).
Time horizon Length of model, usually either life-time of
simulated patients or short-term (e.g. 1 year)
Perspective Either 3rd party payor, large single-payor, or
societal

Treatment effect The magnitude of clinical benefit in terms of

clinical benefit (e.g. reduced rate of
hypoglycemia, reduced HbA1c percent).
These values were usually derived from the
literature – for example by referencing a
single randomised trial, observation study, or
meta-analysis. In the raw data file,
hypoglycemia assumptions were recorded
separately from other clinical outcome
assumptions (i.e. changes in A1c or BMI).
Cost of drug treatment Applicable for both study drug and
comparator drug. The cost source was also
recorded as a separate variable.
Costs associated with increased healthcare These costs may be direct medical costs, for
utilization example, cost of a hospitalisation in the local
context for an acute myocardial infarction
(heart attack) where a patient survived.
Indirect costs such as lost productivity as a
result of illness were generally not reported
in the studies reviewed. Because

Utility weights
Clinical benefit (QALYs)
QALYs gained (or lost)
Cost of complications (excluding drug
therapy or management costs)
Additional (incremental) costs
ICER
Cost-effective?
Cost-effectiveness
threshold (a.k.a. willingness to pay)
Major limitations or concerns
Dominant
complications of diabetes are numerous
(retinopathy, neuropathy, kidney disease,
peripheral vascular disease, and
cardiovascular), individual costs were usually
not recorded, but referenced based upon the
associated table in the full text of the article.
A numeric value assigned to each adverse
event, for example each hypoglycemic event
may be associated with a temporary decrease
in utility for a period of 15 minutes.
Clinical benefit in terms of quality-adjusted
life years gained due to treatment with either
the study drug or comparator, over the time-
horizon of the model
QALYs of analogue minus QALYs of human
Extracted for both human and analogue
insulin products (either directly reported in
article or calculated when possible by
subtracting total costs over time horizon by
total costs of medications and management
costs, when reported separately).
Total costs (pharmacologic therapy,
management, and cost of complications) for
study drug minus total costs for comparator.
Incremental cost effectiveness ratio. In this
table, Additional costs / QALYs gained.
Yes/no; based on whether the ICER is greater
or less than the cost-effectiveness threshold
usually used in the country or geographic
region covered by the study.
Either the reported amount (e.g. 30,000
Euro per QALY) or a fraction of
GDP per capita.
Notes regarding the robustness, accuracy,
reliability or trust-worthiness of the study or
modeling assumptions. Comments regarding
sensitivity analyses or how variable results
were when assumptions varied were also
noted in this data element.
When a new technology or medicine results
in increased QALYs at lower cost than a
comparator technology or medicine.
Policymakers should always select dominant
technologies when given the choice because
these technologies will save money for the
healthcare system while resulting in more
clinical benefit.
Duplicate study selection and extraction was not done because only one individual was
contracted to complete this report.

The quality of CEA studies was determined based upon the data elements described above.
Although no formal checklist was used to assess quality, all of these studies were published in
peer-review journals and generally satisfy good practice in decision-analytic modelling (7,8). For
example, all studies reviewed included appropriate discounting rates (usually three percent) for
both cost and effect outcomes.

Major limitations or concerns with either methodologic quality, choice of comparators,

estimates of comparable clinical benefits/harms derived from literature, estimates of costs and
sensitivity-analyses were noted during the full-text review.

In cost-effectiveness studies, results are typically reported as incremental cost-effectiveness

ratios (ICER), or price per quality-adjusted life-year gained. The prices in the numerator are
usually expressed in local currency or in US Dollars. Based on the ICER and a cost-effectiveness
threshold (if one exists) for a particular country or setting, authors typically conclude whether a
new technology is cost-effective or not cost-effective. In some cases, where the new technology
results in greater QALYs and cost-savings when compared to a comparator, the new technology
may be considered dominant.

2.1.4 Data Analysis

From the raw Excel file compiled above, studies were categorised and summary results were presented
using descriptive statistics for each of 10 domains: geographic scope, sponsor, medicines being
compared, patient population, model type, time-horizon, treatment effects, costs of medicines, costs of
complications (excluding drug costs), and overall statement of cost-effectiveness.

Major limitations to these studies as a whole, how sensitive the results were to changes in
baseline assumptions, as well as their applicability to the value-based framework in Part 2 were
also reported.

2.2 Deriving a Value-based Price for Analogue Insulin

2.2.1 Principles
Using standard principles, concepts and mathematical equations from decision science and
cost-effectiveness analysis, a framework for thinking about a value-based price for analogue
insulin relative to human insulin was constructed. This framework relies upon a fundamental
understanding of value in the assessments of health technologies pioneered by Milton Weinstein
and others, namely that the added value of a new technology is relative to the added (or
incremental) benefit of that technology relative to a previous standard of care treatment divided
by the added (or incremental) cost of that new technology relative to a previous standard of care
treatment (9).

This conception of value is based upon the foundational idea that healthcare resources are in all
cases finite, and that hard choices must often be made by a payer of healthcare services between
one technology versus another. Thus, newer technologies (or treatments) should be adopted
only if they provide better value compared with existing treatments.

In this value-based framework, costs are thus viewed in terms of clinical benefit, with each unit
of benefit being standardized based upon quality-adjusted life-years gained.
2.2.2 Derivation of a Value-based Price for Analogue Insulin
Starting from the basic definition of value described above, a formula was derived to express the
additional, added, or incremental price of analogue insulin relative to the expected clinical
benefit, a cost-effectiveness threshold (aka willingness-to-pay) and expected cost savings
associated using analogue insulin.

This formula does not provide a single value-based price, but rather one which can vary
depending on the inputs selected. For example, less wealthy countries with lower cost-
effectiveness thresholds generally will have a lower value-based price for analogue insulin.

2.2.3 Examples of Value-based Analogue Prices using Data

Extracted from Systematic Review of CEA Studies
The formula derived from the section above was combined with data extracted from the systematic
review to give readers a sample of the range of value-based prices for analogue insulin relative to human
insulin.

Results
3.1 Summary of CEA Studies Comparing Analogue
versus Human Insulin
The full titles of each of the 24 CEA studies can be found in Annex 1 (10-32). The following
sections will describe characteristics of the studies reviewed according to the domains
referenced in the Methods section above.

3.1.1 Geographic Scope

Twenty of the cost-effectiveness studies (83 percent) reviewed in this report were conducted in North
America or Europe. Some studies covered more than one country. Only two CEA studies described
results in settings outside of high-income countries: India, Indonesia, Saudi Arabia (18), and Thailand
(36).
Note: The y-axis indicates the number of CEA studies that report results in each country. The total is greater than
the number of studies reviewed (24) because some studies reported results in multiple countries.

3.1.2 Sponsor

Industry sponsored 22 of the 24 studies (92 percent) reviewed in this report. One study was
sponsored by Health Canada. One study was funded by the National Essential Medicines List of
Thailand. Seventeen studies were sponsored by Novo Nordisk, four by Sanofi, and one by Eli
Lilly and Company.

3.1.3 Medicines Compared VALUE OF INSULIN | 12

Of the 24 studies reviewed, nine (40 percent) evaluated detemir as the new study medicine being
considered. Three evaluated biphasic aspart (13 percent), three evaluated rapid-acting aspart alone (13
percent), five evaluated glargine (22 percent), one evaluated lispro (four percent), and two evaluated
multiple analogue insulins together as a single regimen (nine percent).

Figure 2. Analogue insulins used as study medicine

Among comparator medicines (i.e. human insulins considered standard of care), NPH was the reference
in 13 (57 percent) of studies, regular human insulin five (22 percent), premixed insulin 70/30 was the
reference three times (13 percent) and multiple human insulins twice (nine percent).
Figure 3. Human insulins used as comparator medicine

3.1.4 Diabetes Type

Eleven studies focused on type 2 diabetes (48 percent), eight on type 1 diabetes (35 percent) and
four (17 percent) on both type 1 and type 2 diabetes.

3.1.5 Model Types

IMS’ Center for Outcomes Research Diabetes (CORE) diabetes model was the most frequently
used model in the CEA studies reviewed (17/24 or 71 percent). Microsoft Excel was used to
create models in five of the studies (21 percent). TreeAge was used once and Cardiff Research
Consortium’s UK based Discrete Event Simulation model was used once.

The CORE diabetes model has been validated against published studies with a predictive abilities of
0.8861 to 0.9778, providing a reasonably accurate representation of real-life settings (16). Briefly, these
computer simulations are economics models that predict the development of various health states for
individual patients or a cohort of patients based upon baseline characteristics, assigned treatment
regimen (analogue or human insulin) and probabilities of developing an adverse event or diabetes-
related complication based on sub-model event equations (Figure 6) (33, 34). Baseline characteristics
were usually obtained from a single randomised trial, observational study. Reductions in the rate of
complications were usually due to presumed reductions in HbA1c obtained from short term trials or
observational studies (typical duration less than 24 weeks) or less commonly from a meta-analysis of
published trials. These differences were assumed to persist for the entire duration of the model
simulation. Similiar assumptions were made for reductions in the rate of hypoglycemia.

3.1.6 Time Horizon

A majority of studies (18 of 24, or 75 percent) used a life-time time horizon. This is consistent
with conventional practice as well as guideline recommendations regarding length of time to
calculate costs and benefits. Five studies used time horizons that were either short (one to three
years) or medium term (five to 10 years). The stated goals of these shorter studies were to assess
short term value for money associated with newer analogue insulin products.

3.1.7 Treatment Effects

The CEA studies reviewed in this report used a variety of sources to estimate the effect on risk of
hypoglycemia and changes to HbA1c and weight (sometimes referred to body mass index(BMI)).
For example, in the 16 studies reporting differences in outcomes among people with type 2
diabetes, a wide-range of assumptions were used in the cost-effectiveness models. For
hypoglycemia, the effect of analogue insulin compared to human insulin ranged from no
difference (three studies) to a relative reduction in the event rate of 80 percent. Studies also
differed with respect to types of hypoglycemia. Many studies assumed that mild hypoglycemia
events which are usually more frequent would be reduced at a different rate than major or
severe events (often defined as needing the assistance of one other person). Baseline rates also
varied widely. For example, some studies assumed a baseline rate of 286 per 100 person years to
a high of 1638 events per 100 person years. Assumed baseline rates of major hypoglycemia was
less variable, in part since these events happen much less frequently (Table 3). For HbA1c,
treatment effects also varied widely, from a low of -.05 percent to a high of 2.1 percent reduction
being attributable to analogue insulin when compared against human insulin. For BMI, the
estimated treatment effect of analogue versus human insulin ranged from -0.18 to -0.38.

Many of the sources used to generate treatment effects seen in Tables 2 and 3 were of low to moderate
quality. For example, one study based its treatment assumptions on a single 24 week observational study
with unlikely results (two percent decrease in HbA1c) extending into 30 year clinical benefits (18).
Another study based its assumptions on a single observational study of general practitioners in the UK
with methodological flaws (e.g. prevalent user design, immortal time bias, suboptimal confounding
control) and unlikely results given the short timeframe of the study. For example, the referenced study
calculated an adjusted hazards ratio for acute myocardial infarction of 0.69 and stroke of 0.58 (21). One
study drew its assumptions from a subgroup analysis of a short term, large, single-arm cohort study
(IMPROVE) (25). It is highly difficult, if not impossible to draw meaningful clinical conclusions about
comparative treatment benefits from single-arm observational studies because of
pharmacoepidemiologic biases including confounding by indication.

It is also important to note that the overall evidence supporting the clinical superiority of analogue
insulins in type 2 diabetes is relatively weak. For example, several direct head-to-head trials and meta-
analyses have shown only modest benefit (almost exclusively shown with respects to the rates of
hypoglycemia). Most parallel arm, comparative trials utilising random allocation of treatment assignment
has not shown any benefits in favor of analogue insulins with regards to HbA1c or hard clinical endpoints
(19,35).

Table 2. comparative effect estimates for short term clinical outcomes (hypoglycemia, Aic, weight/BMI)
in CEA studies among people with type 1 diabetes

Lead Author Active Drug Comparator Effect on Effect on Effect on

Drug hypoglycemi HbA1c Weight or
a (risk ratio BMI
or reduction
in absolute
rate per 100
person-
years)
Long-acting insulin analogues
Morales detemir NPH RR 0.84 n/a +0.91kg
Grima glargine NPH Not described -0.4%
Valentine detemir NPH (and Hypo risk -0.22% Body weight
glargine) reduced by reduced 1kg at
21% 18 weeks.

In type 1 diabetes, a balanced view of the literature (outside of the CEA studies reviewed here)
suggests a slightly different conclusion. One large meta-analysis of 27 RCTs including 7496
patients concluded that long-acting analogue insulins were modestly superior to intermediate-
acting human insulins with respects to HbA1c, hypoglycemia and weight (mean A1c difference
-0.39 percent, risk of severe hypoglycemia OR 0.62) (5).

3.1.8 Costs of Analogue and Human Insulins

Most of the analogue insulins in this review were priced at a level that was between 10 percent to
100 percent higher price than human insulin products (range 0 percent to 1141 percent). In one
study, the cost of analogue insulin was the same as the cost of human insulin, in part due to a
governmental decision not to reimburse rapid-acting insulin analogues at a higher rate than
regular human insulin (21). In another study, lispro only cost GBP 187 more than regular human
insulin over the lifetime of an individual (or ~ GBP 3.74 per year over 50 years) (28). This is
likely due to the fact that these two countries (Germany and the UK, respectively) have strong
health technology assessment agencies, large purchasing power and other price control
mechanisms.

On the other end of the spectrum, three studies assumed that the price of analogue insulin was
substantially higher than human insulin. For example, one study using a US perspective
assumed that biphasic aspart was 160 percent higher price than human insulin, although it is
unclear whether the prices quoted were list, reimbursement, or net prices since the source was
not cited in the article (25). Another conducted from the perspective of Canada assumed that
detemir 243 percent higher price than NPH in the setting of type 2 diabetes (140 percent more
expensive in the type 1 setting, due to different dosing assumptions) (10). These results are not
surprising when viewed in context: both the US and Canada lack an efficient mechanism at the
national level to control drug prices. Canada in general has lower-brand name drug prices, but
there may be substantial variation from province to province. Finally, a Thai study estimated
that the annual price per patient for insulin glargine was 1242 percent the cost of NPH insulin
(7380 THB vs 594 THB).
Prices of analogue and human insulin products varied widely depending on study and location of the CEA
studies reviewed (Table 3). Furthermore, there was little standardisation in how pharmacologic or drug
management costs or prices were reported. Some studies reported prices per international unit of
insulin or per dispensed unit (i.e. 1 vial or pack of 5 pens). Other studies reported costs per day or per
year of treatment. Still others reported costs over the entire time-horizon of the study, in some cases
over 30 years.

Table 4. Comparing the cost of analogue and human insulin among selected CEA studies* by type of
insulin (long acting, pre-mixed, rapid acting or multiple insulin types)

3.1.9 Costs of Diabetes-related Complications (excluding

medicine costs)
Table 5 reports the costs of medical complications attributable to a simulated cohort of people
living with type 1 or type 2 diabetes, being assigned treatment with either analogue or human
insulin, over the stated time-horizon. Costs of diabetes-related complications may be either
direct or indirect. An example of a direct medical cost is hospitalisation in the local context for
an acute myocardial infarction (heart attack) where the person survived. An example of an
indirect cost is lost productivity as a result of illness. Indirect costs were generally not reported
in the published CEA studies reviewed. Because complications of diabetes are numerous
(retinopathy, neuropathy, kidney disease, peripheral vascular disease, and cardiovascular),
individual costs are not separately reported in the summary table below, but can be found in
full-text of each reviewed article (see Annex 1).

The costs of complications here also specifically exclude the costs of medications, because the
empirically derived value-based pricing formula in Part 2 of this report relies on the additional or
incremental cost of complications attributable to using human insulin, excluding medicine treatment
costs.

Table 5. cost of complications (excluding drug costs) in selected CEA studies

3.1.10 Summary of Cost-effectiveness Results

While 21 out of 24 (88%) CEA studies found that analogue insulins were cost-effective when
compared to human insulins (Table 6) the following three issues must be taken into
consideration:

1. Industry sponsorship is a strong predictor of claimed cost-effectiveness

There was a high correlation between the funding source of the CEA study and determination of
cost-effectiveness. Only two of the published CEA studies in this review were funded by
governmental bodies (i.e. no industry funding) (16,36). In both studies, the analogue(s) being
considered were found to be either not cost-effective (in the case of Thailand), or mostly not
cost-effective (in the case of Canada). In the study funded by Health Canada, rapid-acting
insulins aspart and lispro were found to be cost-effective for people with type 1 but not people
with type 2 diabetes (16). In the study funded by the National Essential Medicines List of
Thailand, the long-acting insulin analogue glargine was not cost-effective compared to NPH for
use in people with type 2 diabetes, mostly due to high glargine costs (over 1200 percent more
expensive than the comparator) and a low cost-effectiveness threshold (US$4714.2/QALY). In
the Thai study, once glargine costs were reduced by more than 43 percent compared against the
base-case scenario, glargine became a cost-effective option. One study found mixed results:
aspart was cost effective in three of four. European countries examined, but not cost-effective in
Poland (14). Overall, analogue insulins were found to be dominant in six studies (26 percent),
meaning that they resulted in more QALYs gained at less cost when compared against human
insulin.

2. Cost-effectiveness is related to the comparative price of the analogue insulin

Rapid-acting analogues

The rapid-acting analogues aspart and lispro were found to be dominant compared to regular
human insulin in four out of five CEA studies. In the fifth study by Palmer and others (2008),
aspart was dominant when compared against regular insulin in Sweden and Spain, cost-effective
in Italy, but not cost-effective in Poland. The most likely reason these products were considered
dominant is because of the low medication costs when compared against regular human insulin.
For example, in three of the country settings examined (Germany, Japan and the UK), prices for
the rapid-acting analogue were only zero percent to three percent more expensive than regular
human insulin. This suggests that when prices are comparable between human and analogue
products, the probability of cost-savings to the health system or society is high.

Long-acting analogues
In contrast, the CEA studies that examined the long-acting analogues glargine or detemir
exhibited large variations in comparative price (range: 14 percent to 1141 percent more
expensive than NPH). In the two studies where the price of the analogue was only slightly higher
than NPH (i.e. 14 percent to 36 percent more expensive), the analogues were found to be
dominant, or cost-saving. In the seven studies where the price of the long-acting analogue was
found to be cost-effective, the price of the long-acting analogue was between 16 percent and 243
percent more expensive than the human insulin (average 98 percent more expensive). On the
higher end of this range (between 66 percent and 243 percent more expensive), the long-acting
analogue was still cost-effective but was associated with a higher ICER value. For example, in
Ridderstrale (2013) and Tunis (2009), glargine and detemir had ICER’s of €25,097 per QALY
and $CAD 24,389 per QALY, respectively when compared against NPH.

3. Cost-effectiveness of type 1 versus type 2 diabetes

Twelve studies addressed type 1 diabetes, while 16 addressed type 2 diabetes (note: four studies
overlapped because they assessed cost-effectiveness among people with type 1 and type 2
diabetes).

In type 1 diabetes, all but 1 study was funded by industry. All were conducted in high-income
settings (Annex 2). Two studies found that the analogue insulin was dominant (lispro in the UK
and glargine in Sweden). Nine studies found that analogue insulin was cost-effective, under
various cost-effectiveness thresholds meant for high-income settings. The final study was
funded by Health Canada and found heterogenous results: rapid-acting analogues were
determined to be cost-effective, with ICERs in the range of $CAD 28,996 (aspart) to dominant
(lispro). However, long-acting analogues was found to be not cost-effective in this Canadian
study, with ICER’s ranging from $CAD 87,932 (glargine) to a staggering $CAD 387,729
(detemir).
In the 16 studies addressing type 2 diabetes (or both type 2 and type 1 diabetes), all but two
studies were funded by industry. Among type 2 diabetes studies, four studies reported that
analogue insulin was dominant when compared against human insulin. Nine studies found that
the newer analogue was cost-effective, again, under various cost-effectiveness thresholds meant
for high-income settings. Two studies reported heterogeneous results: for example, the Palmer
study examined cost-effectiveness of aspart compared against regular human insulin in four
European countries. In that study, aspart was found to be dominant in Sweden and Spain, cost-
effective in Italy (based on a CE threshold of €30,000 per QALY), but not cost-effective in
Poland (ICER €290,486). The Health Canada funded study reported that aspart was cost-
effective when compared against regular human insulin, however lispro (ICER $CAD 130,865)
and glargine were not (CAD$ 387,829). Determir was dominated by NPH, meaning that it was
found to result in greater costs and lower QALYs. The study funded by the Ministry of Health of
Thailand, found that glargine was not cost-effective when compared against NPH.

Based on these findings, it is difficult to make any firm recommendations regarding the cost-
effectiveness of analogue insulins comparing type 1 versus type 2 diabetes. In both cases, while
more studies found that analogue insulins were cost-effective than not, the results were
heterogeneous. Furthermore, nearly all studies were funded by industry and conducted in high-
income settings.

Therefore, cost-effectiveness results seem to be driven more by the funding sponsor,

assumptions made with respect to clinical effectiveness / safety, and the comparative price of
analogue insulin, than by the type of diabetes being treated (i.e. type 1 versus type 2 diabetes).

Recommendations

Policymakers in LMICs should choose long-acting analogues only when their price(s) are
comparable or slightly higher price than human insulins (i.e. NPH and 70/30) because
available data from published modeling studies suggest that the cost-effectiveness of
ananalogue insulin is sensitive to its procured price. Long-acting analogue insulins are highly
unlikely to be cost effective in LMICs when their price exceeds the price of human insulin.

Note that this general recommendation is based on a limited dataset, where over 90 percent of
published CEA studies were funded by industry where the risk of bias (in favor of cost-
effectiveness for the newer, more expensive product) is high. If these studies are excluded, the
long-acting analogues are not cost-effective when compared against NPH. In the Thai study,
glargine was not found to be cost-effective compared to NPH in people with type 2 diabetes, in
part because it was over 12-fold more expensive. In the study funded by Health Canada, neither
glargine nor detemir were cost effective when compared against NPH for either type 1 or type 2
diabetes. Most studies were undertaken in high income countries.
Policymakers in LMICs should choose rapid-acting analogues only for patients with type 2
diabetes when the price difference between the analogue and human insulin is negligible.

With respect to rapid-acting or pre-mixed insulin analogue products in type 2 diabetes, the data is even
more limited (in part due to fewer studies). However, based on analysis above, policymakers should
choose rapid-acting analogues when their price difference between rapid-acting analogues and regular
human insulin is between zero percent to three percent. Note that this recommendation is based on
limited CEA studies mostly conducted in higher-income settings where the modeled price of the rapid-
acting analogue insulin was almost the same price as regular human insulin. In these settings, studies
found that rapid-acting analogues were dominant or cost-effective, representing either overall cost-
savings or good value for money.

3.1.11 Discussion

This systematic review examined 24 published CEA studies comparing an insulin analogue
against a human insulin product for type 1 or type 2 diabetes. We found that the determination
of cost-effectiveness for an insulin product is related to both the funding source and the
comparative price of the analogue insulin product. Unsurprisingly, industry funded studies
almost universally concluded that analogue products were cost-effective. The two independently
funded studies had differing conclusions: insulin analogues were found to be generally not cost-
effective when compared against human insulin (with a possible exception being rapid-acting
analogues for type 1 diabetes). Additionally, we found a relationship between the price of an
analogue and its likelihood of being cost-effective. Based on our analysis of comparative insulin
prices and cost-effectiveness results from published studies, we propose two recommendations
for policymakers:

1) Policymakers in LMICs should choose long-acting analogues only when their price(s) are
comparable or slightly higher priced than human insulins (i.e. NPH and 70/30) because
available data from published modeling studies suggest that the cost-effectiveness of an
analogue insulin is sensitive to its procured price. Long-acting analogue insulins are highly
unlikely to be cost effective in LMICs when their price exceeds the price of human insulin.

2) Policymakers in LMICs should choose rapid-acting analogues only for people with type 2
diabetes when the price difference between the analogue and human insulin is negligible.

There are several limitations to our review. A major limitation was the fact that over 90 percent
of published CEA studies examined were sponsored by industry. Second, this author was only
able to identify two CEA studies from LMIC settings. This is particularly relevant because much
of the global burden of insulin-dependent diabetes is bourne by people living in LMICs and
because of the disproportionate amount of healthcare spending that is dedicated to these
patients. Therefore, this review suggests that more independent studies focused on LMIC
settings are urgently needed.

Finally, it should be noted that the overall conclusions of this review were based on ICER’s and cost-
effectiveness thresholds derived from a variety of different payor and geographic perspectives. In many
cases, when study authors varied their assumptions in sensitivity analysis, conclusions with regards to
cost-effectiveness changed. For example, in one study, when no difference in rates of severe
hypoglycemia comparing lispro against regular human insulin was tested in a sensitivity analysis,
probability of being cost effective went from 84 to 59 percent (28).

3.2 A Value-based Price for Analogue Insulin

3.2.1 Derivation of Value-based Formula for Pricing Analogue
Insulin

Value (V) is generally defined as change in clinical benefit (B), expressed in QALYs divided by
change in cost (C), in some arbitrary currency amount (usually $).
V=ΔB/ΔC

Rearranging yields:
ΔC = ΔB / V

Value is also frequently defined as the willingness of a society to pay (in currency amounts) per
QALY gained or saved. For unit cancellation purposes (i.e. to get rid of QALYs), let us assume
that the inverse is also true, that value can be expressed as the inverse of the willingness-to-pay
(i.e. V = QALY / WTP).

By substitution:
ΔC = ΔB * WTP per QALY

The WHO has recommended that the williness-to-pay (or cost-effectiveness threshold) be set at
1 to 3x GDP per capita:

ΔC = ΔB * (1x to 3x GDPpc)

ΔC can be expressed as the difference in total cost between analogue (C a) and human insulin (Ch):
ΔC = Ca - Ch
Therefore:

Ca - Ch = ΔB * (1x to 3x GDPpc)
Since total cost is the sum of the cost of the medication and the cost of the complications
(C(complic)) as a result of being assigned one medication versus another:

Ca - Ch = (Ca (Rx) + Ca (complic)) – (Ch (Rx) + Ch (complic))

Or:
Ca (Rx) + Ca (complic)) - Ch (Rx) - Ch (complic)

Rearrangement yields:

Ca (Rx) - Ch (Rx) + Ca (complic)) - Ch (complic)

Therefore:

Ca (Rx) - Ch (Rx) + Ca (complic)) - Ch (complic) = ΔB * (1x to 3x GDPpc)

Finally, solving for Ca (Rx) and Ch (Rx) yields:
Ca (Rx) - Ch (Rx) = ΔB * (1x to 3x GDPpc) - Ca (complic) + Ch (complic)
Or:
Ca (Rx) - Ch (Rx) = ΔB * (1x to 3x GDPpc) + Ch (complic) - Ca (complic)

3.2.2 Value-based Analogue Insulin Price Expressed in Words

Stated in words, the additional cost (price) of analogue insulin over the price of human insulin is
equal to the expected clinical benefit multiplied by 1x to 3x GDP per capita plus the difference in
expected costs attributable to diabetes related complications between being treated with human
and analogue insulin.
In other words, the value-based price of analogue insulin depends on both the expected
additional clinical benefit (in QALYs), the willingness to pay of a given country or health system,
and the expected cost savings with respects to a reduced rate of diabetes related complications.
Since both the expected additional benefit of analogue insulin compared to human insulin and
the reduced costs of diabetes related complications can be empirically obtained from published
sources, it is possible to calculate a relative price of analogue insulin that reflects its marginal
benefits.

The calculated additional price can be interpreted and used in the following manner: if the price
of the analogue insulin being considered is higher than the price that reflects its marginal value
(based upon commonly accepted cost-effectiveness thresholds of 1x to 3x GDPpc), then that
analogue insulin represents poor value for money. If the price of the analogue insulin being
considered is less than the calculated value-based price, then that price may represent good
value for money.

3.2.3 Some Real-world Examples of a Value-based Analogue

Price
3.2.3.1 An Example Using Data from a Study of People with Type 2 Diabetes in
Germany
Let us assume that a lower-middle income country with GDP per capita of €2000 is interested in
providing short or rapid-acting insulin for its people with type 2 diabetes. Should it purchase aspart (a
rapid acting insulin analogue) or regular human insulin? At what price would a procurement officer be
indifferent to choosing between analogue or regular human insulin?

One can use the value-based formula derived above and data from the Liebl et al 2014 study in
Germany to approximate the price at which one should be indifferent between these two
products.21 This study assumed unlikely treatment benefits. For example, aspart was associated
with a reduction in macrovascular events of 15 percent after only a mean follow-up time of 3.5
years, based on the results of a single observational study conducted in 1900+ general practices
in Germany. Even with these unrealistic assumptions, the article reported a mean benefit of only
0.02 QALYs over three years as a result of choosing aspart. At the same time, non-medication
costs associated with complications were €4417 and €5973, for analogue and human insulin
respectively. Using the formula derived above, the marginal price at which one should be
indifferent between choosing between analogue and human insulin would be when aspart was
priced €532 to €559 per year more than human insulin, or an additional €44 to €47 per
month.

In this example, the costs of complications over three years is relatively high, in part because the
cost of healthcare services in Germany is higher than it would be in most LMICs. These assumed
or expected complication costs are probably lower in countries that are less wealthy when
compared against Germany. If the costs of complications were assumed to be reduced by 75
percent for both analogue and human insulins, the additional price for aspart at which one
would be indifferent between choosing that product or human insulin would fall to only and
additional €12 to €14 per month.

3.2.3.2 An Example Using Data from a Study of People with Type 2 Diabetes in
India
In Gupta et al 2015, a pre-mixed insulin analogue containing biphasic aspart (biAsp30) resulted in 2.5
additional QALYs over 30 years when compared against premixed insulin 70/30 (human) based on clinical
results from a single 24-week observational study. Over the same time frame, complication costs were
reduced by US$482.18 Using the value-based formula above and a GDPpc of US$1582, the price of
biphasic aspart reflecting its marginal benefits over 70/30 would be an additional US$12 to US$34 per
month. Under more realistic scenarios, where expected QALY gains would be reduced, the marginal price
at which one would be indifferent between this analogue insulin and a comparable human insulin would
also fall. For example, if QALY gains were reduced by 50 percent, the additional price by which one would
be indifferent to biAsp30 versus premixed 70/30 would be: US$6.8 to US$17.8 per month.

3.2.4. Summary of Value-based Price for Analogue Insulin

A value-based price for analogue insulin can be derived that expresses the additional price of
analogue insulin (over human insulin) as a function of the expected clinical benefit in QALYs,
expected costs of diabetes-related complications and a country’s willingness-to-pay
(conventionally valued at 1x to 3x a country’s GDPpc).

Ca (Rx) - Ch (Rx) = ΔB * (1x to 3x GDPpc) + Ch (complic) - Ca (complic)

In words, the additional cost (price) of analogue insulin over the cost of human insulin is equal
to the expected clinical benefit multiplied by (one to three times) GDP per capita plus the
difference in expected costs attributable to diabetes related complications between being treated
with human and analogue insulin. Clinical benefit and costs attributable to diabetes related
complications can either be estimated de novo, in countries that have the capacity to do so, or
estimates can be obtained from the published literature, as shown in two examples in this
report.

An advantage of using published estimates is that CEA studies discount both future benefits and
costs. A limitation of this approach is that extrapolating complication costs from one setting may
not accurately represent costs in other settings. For example, as shown in the two highlighted
examples, the additional medication price is highly dependent on both the estimated clinical
benefit as a result of choosing the analogue insulin and the differential costs of diabetes-related
complications (e.g. less hospitalisations and management costs associated with reduced rates of
hypoglycemia). Another limitation is that this equation only allows a decision-maker to estimate
the additional price of analogue insulin that reflects its marginal benefits over the current price
of human insulin. Thus, the resulting price calculated is on a relative scale, not an absolute scale.
Therefore, changes in the current price of human insulin will affect judgements about the value
of an analogue insulin being considered.

Finally, this approach should be viewed as a “quick-and-dirty” way of establishing a ceiling price for
analogue insulin that represents poor value for money. It may also be used as a tool to help in insulin
price negotiations with manufacturers or suppliers. This approach should not be used as a single decision
rule that triggers procurement of analogue insulins. As acknowledged by several authors within the
WHO, decisions about funding new health interventions should not be based on any single number or
threshold (37). Rather, these decisions should incorporate other important priorities such as: the clinical
or policy context, the overall budget impact, the opportunity cost of forgoing other treatments with the
same limited resources, and finally fairness/equity concerns. Furthermore, value-based calculations can
be misused by the industry to price new drugs or treatments in various jurisdictions as a profit-
maximising strategy.