SCIENTIFIC CONTRIBUTION

Complexity and indeterminism of evidence-based public health: an analytical framework

Francesco Attena

Published online: 16 March 2014

� Springer Science+Business Media Dordrecht 2014

Abstract Improving the evidence in public health is an

important goal for the health promotion community. With

better evidence, health professionals can make better

decisions to achieve effectiveness in their interventions.

The relative failure of such evidence in public health is

well-known, and it is due to several factors. Briefly, from

an epistemological point of view, it is not easy to develop

evidence-based public health because public health inter-

ventions are highly complex and indeterminate. This paper

proposes an analytical explanation of the complexity and

indeterminacy of public health interventions in terms of 12

points. Public health interventions are considered as a

causal chain constituted by three elements (intervention,

risk factor, and disease) and two levels of evaluation (risk

factor and disease). Public health interventions thus differ

from clinical interventions, which comprise two causal

elements and one level of evaluation. From the two levels

of evaluation, we suggest a classification of evidence into

four typologies: evidence of both relations; evidence of the

second (disease) but not of the first (risk factor) relation;

evidence of the first but not of the second relation; and no

evidence of either relation. In addition, a grading of inde-

terminacy of public health interventions is introduced. This

theoretical point of view could be useful for public health

professionals to better define and classify the public health

interventions before acting.

Keywords Causality � Complexity � Epidemiology � Epistemology � Public health

Introduction

Public health interventions, complexity, and indeterminism

are strictly interrelated concepts. Briefly, public health

interventions are considered health and epidemiological

activities of high complexity. The main epistemological

feature of these complex systems is their indeterminacy;

the indeterminism of complex systems represents a strong

limitation to the claim of evidence-based public health.

Evidence-based public health (EBPH)

Improving the evidence in public health is an important

goal for the health promotion community. With better

evidence, health professionals can make better decisions

to achieve effectiveness in their interventions. EBPH is

‘‘the conscientious, explicit, and judicious use of current

best evidence in making decisions about the care of

communities and populations in the domain of health

protection, disease prevention, health maintenance and

improvement’’ (Jenicek 1997). In more recent years, the

perspectives of community members have been included,

which has helped foster a more population-centered

approach (Brownson et al. 2009); thus, EBPH has also

become ‘‘a process of integrating science-based inter-

ventions with community preferences to improve the

health of populations’’ (Kohatsu et al. 2004). Some crit-

ical points related to EBPH are as follows: What kind of

evidence is it possible to produce with regard to both

observational and experimental studies? How should this

evidence be graded (Tang et al. 2008)? How should this

evidence be translated into public health practice and

have a bearing on the decision-making process (Rychetnik

et al. 2012)? And how should this whole process of

EBPH be constructed (Briss et al. 2000)?

F. Attena (&) Second University of Naples, Naples, Italy

e-mail: francesco.attena@unina2.it

123

Med Health Care and Philos (2014) 17:459–465

DOI 10.1007/s11019-014-9554-0

Complexity

The concept of complexity is elusive and with uncertain

boundaries. It can be defined as ‘‘a scientific theory which

asserts that some systems display behavioral phenomena

that are completely inexplicable by any conventional ana-

lysis of the systems’ constituent parts’’ (Hawe et al. 2004).

More particularly, the complex systems ‘‘are highly com-

posite ones, built up from very large numbers of mutually

interacting subunits (that are often composites themselves)

whose repeated interactions result in rich, collective

behavior that feeds back into the behavior of the individual

parts’’ (Rickles et al. 2007).

For the purpose of this paper, we consider the following

characteristics of complex systems (Plsek and Greenhalgh

2001; Pearce and Merletti 2006; Galea et al. 2010;

Tremblay and Richard 2011): the large number of inter-

acting components; self-organization; circular causality or

feedback; and emergent properties. The indeterminism of

complex systems and their unpredictability are the episte-

mological consequences of these properties.

Complex systems differ from complicated systems. The

latter also have many interacting components, but they

interact in a mechanistic manner (mechanical causality), in

which the whole is equal to the sum of their parts and

where the rules are linear causality and predictability.

Computers and airplanes are very complicated—but not

complex—systems because they must have strictly pre-

dictable behaviors. Therefore, whereas public health

interventions are closer to complex systems, clinical

interventions are less complex, having characteristics more

similar to those of complicated systems (Attena 1999).

Clinical interventions might be also more complicated than

public health interventions, but they must be less complex

(i.e. more binding, deterministic and predictive) because

they deal with ill people who need, with a high degree of

certainty, to heal or improve their disease status. Public

health interventions are less bound to this requirement, and

they can afford a greater degree of error.

Indeterminism and unpredictability

As previously shown, if indeterminism is an epistemolog-

ical consequence of complexity, and public health inter-

ventions are highly complex systems, then public health

interventions involve a high level of indeterminacy. The

indeterminism of public health interventions may be

interpreted in two ways. According to epistemic indeter-

minism, seeming indeterminism is only a consequence of

the lack of knowledge about an underlying determinism

(chance due to ignorance). Ontological indeterminism, in

contrast, describes indeterminism as a real, ontological

characteristic of nature (chance lies in the nature of things).

After a long period of dominance by the deterministic point

of view beginning with Laplace’s demon, ontological

indeterminism became dominant among the physics com-

munity in relation to quantum theory. In the field of life

sciences, in contrast, epistemic indeterminism remained

implicitly dominant. Epistemic indeterminism may be

included in the wider conception of general determinism,

which states that ‘‘everything is determined in accordance

with laws by something else’’ (Bunge 1979). Owing to

complexity, the indeterminism of complex systems—and

consequently of public health interventions—is certainly

irreducible in principle. However, though it is very difficult

to determine whether the indeterminism is epistemic or

ontological, here the paradigm of epistemic indeterminism

is sustained.

Public health interventions involve a higher degree of

indeterminism than do clinical interventions. Furthermore,

for a given degree of indeterminism, the outcomes of

public health interventions present a given degree of

unpredictability, with scientific prediction defined as the

‘‘deduction of propositions concerning as yet unknown or

unexperienced facts, on the basis of general laws and of

items of specific information’’ (Bunge 1979). Moreover, if

the indeterminism is epistemic, even predictability has

epistemic significance, so it may be gradually reduced by

repeating experiments and improving methodologies.

To this point, these considerations about the controversy

between determinism and indeterminism have very much

concerned risk factor epidemiology (Susser 1973; Kar-

hausen 2000; Parascandola and Weed 2001; Olsen 2003;

Parascandola 2011).

This paper presents an analytical explanation of the

complexity, and consequently the indeterminism and the

unpredictability, that underlie public health interventions.

Then, starting from the characteristics of public health

interventions, four typologies of evidence are discussed.

Complexity of public health interventions

Public health interventions are intended to promote health

or prevent ill health in communities or populations (Ry-

chetnik et al. 2002). They are distinguished from clinical

interventions (mainly clinical trials), which are intended to

treat groups of ill people. Public health interventions

include a wide range of activities: policies, laws, and reg-

ulations; organizational or community developments;

education of individuals and communities; engineering and

technical developments; service development and delivery;

and communication, including social marketing (Rychetnik

et al. 2004). More synthetically, public health interventions

can be divided into three broad categories: clinical;

behavioral; and environmental.

460 F. Attena

123

Clinical prevention interventions are those conveyed by

health-care providers, often within a clinical setting (e.g.,

vaccines), and interventions using some kind of drug for

preventive purposes (e.g., statins in primary prevention).

They differ from clinical interventions because they

involve healthy people. Behavioral strategies include

health-promotion interventions, in which people are moti-

vated to modify unhealthy behavior (e.g., stopping smok-

ing). Environmental interventions are those that society can

impose by acting on the environment (e.g., water purifi-

cation) (Haddix et al. 1996). When a public health inter-

vention provides an evaluation of outcomes, we enter the

field of observational or experimental epidemiology. Both

within the context of observational and experimental epi-

demiology, the characteristics of public health interven-

tions with respect to clinical interventions—and

particularly with respect to randomized controlled trials

(RCTs)—are reported below and synthesized in Table 1.

Here, we largely make an evaluation of health-promotion

interventions, though sometimes also of environmental

ones. Clinical prevention interventions, which are similar

to clinical interventions, have been excluded.

Length of the causal chain

Theoretically, and with the exclusion of confounding fac-

tors, in clinical trials the causal chain is constituted by two

elements: the treatment or clinical intervention (cause) and

the outcome (effect). Sometimes, so-called intermediate

variables are considered and evaluated, such that the chain

can appear to comprise three elements. By contrast, in

public health interventions, the causal chain is fundamen-

tally constituted by three elements: the preventive inter-

vention, the risk factor, and the outcome. The intervention

(cause) acts on the risk factor (effect), and the risk factor

(cause) acts on the outcome (effect). To evaluate the

effectiveness of the intervention, it is possible to examine

the first and/or the second relation (Fig. 1).

Dual outcomes

The consequence is that the public health interventions

have two levels of evaluation—reduction of the risk factor,

and reduction of the corresponding disease. For example,

in a health-promotion intervention to reduce childhood

obesity by modifying the diet, evaluation can involve either

the intervention ability to modify the diet (first level) and/

or the reduction in obesity (second level). The choice to do

neither, one, or both evaluations depends on several fac-

tors, such as the degree of context dependence, the reli-

ability of causal relations, available resources, and whether

any experimental setting has been designed.

Weakness of the causal chain

Often despite extensive research efforts, both causal rela-

tions can be uncertain: the first one simply because of the

characteristics of indeterminacy and unpredictability of

public health interventions outlined above; the second one

for the well-known limitations of risk factor epidemiology.

Starting from such issues, four typologies of evidence in

public health are presented below.

Role of other confounding/risk factors

When a public health intervention operates on a single risk

factor of disease, such as air pollution for lung cancer and

diet for obesity, we know that other risk factors also act on

the disease. Therefore, an intervention to reduce the inci-

dence of a disease can work on one or more risk factors.

For example, to reduce the incidence of cardiovascular

diseases, it is possible to achieve effects simultaneously,

such as through health education about smoking, diet, and

sedentary habits.

Role of other diseases

Likewise, when it is the intention of a public health

intervention to reduce one or more risk factors of a disease,

it can also act on other diseases that have the same risk

factor or factors. An intervention directed at smoking, diet,

and sedentary habits can also influence other diseases.

Table 1 Characteristics of public health interventions

1. Length of the causal chain

2. Dual outcomes

3. Weakness of the causal chain

4. Role of other confounding/risk factors

5. Role of other diseases

6. Plurality of interventions

7. Context dependence

8. Words as causal factors

9. Low incidence of the outcome and length of the observation

period

10. Difficulty in statistical analysis

11. Difficulty in obtaining compliance from the study population

12. Difficulty in applying the RCT design

Fig. 1 Causal chains in clinical and public health interventions

Complexity and indeterminism of evidence-based public health 461

123

Plurality of intervention

Finally, to reduce a risk factor with greater efficacy, it is

possible to achieve simultaneous acts. The fight against

smoking includes health-promotion interventions, banning

smoking in public places, and increasing the price of ciga-

rettes. Figure 2 presents a simplified model and an example of

the complex interaction of causes and effects. Other available

causal models, such as the traditional web of causation

(MacMahon and Pugh 1996) and causal diagrams (Joffe et al.

2012), do not address the whole web or chain that starts with

one or more public health interventions. Thus, though the web

of causation is a finite, definite model, because it considers

only one disease, this model can be extended to infinity for the

extensibility of all three causal elements.

Context dependence

This characteristic has been extensively investigated (Pickett

and Pearl 2001; Dobrow et al. 2004; Jackson et al. 2005;

Kemm 2006) because public health interventions work in an

environment in which social, cultural, economic, and political

factors—as well as the competence of the operators—interact.

Accordingly, it is very difficult to apply the principles of

predictability and repeatability. In other words, and unlike the

case with clinical trials, the request to yield the same result

wherever and whenever they are carried out cannot be met. Of

course, within public health interventions, health-promotion

interventions are the most context dependent—mostly

because words, not drugs or equipment, are employed.

Words as causal factors

Unlike clinical interventions and other public health

interventions, health-promotion interventions, i.e., health

education, use words as causal factors to reduce risk fac-

tors. Thus, the first relation of the causal chain is funda-

mentally a bidirectional relationship that involves a deep

interaction between delivering and receiving the interven-

tion. In a broader sense, we have entered the realm of soft

science, such as psychology and sociology, where the issue

of determinism/indeterminism is still more difficult to deal

with and where the indeterminism (ontological or episte-

mological) of human actions and interactions is deeper than

in hard science.

When we need to assess the effectiveness on the disease

outcome rather than on risk factor reduction, an additional

three well-known items of more complexity are implicated.

Low incidence of the outcome and length

of the observation period

Though in clinical trials, a high incidence of outcome is

expected over a short period, the incidence of diseases to

be prevented in public health is much lower and is over the

long term. The major consequence, in the longitudinal

studies, is the need for a very high number of participants

and a very long period of observation.

Difficulty in statistical analysis

In general, following the definition used in this paper for

complexity, statistical models are (very) complicated—not

complex—systems, as they are deterministic and predict-

able. Indeed, if we repeat the same (very) complicated sta-

tistical calculation starting from the same initial conditions,

we always obtain the same result (predictability). Moreover,

the difficulty of applying and interpreting statistical models

concerns both public health interventions and clinical

interventions. However, for two reasons there is somewhat

more difficulty when public health interventions are con-

cerned. First, the more substantial role of confounders in risk

factor epidemiology makes it more difficult to establish

causal association. Second, the outcomes of public health

interventions yield, generally, only small differences

between the treated group and the control group, making it

difficult to obtain statistical significance (Buring 2002).

Difficulty in obtaining compliance from the study

population

It is well known that the length of the observational period

and the magnitude of the sample size are related to

increased difficulty in obtaining compliance and to a high

risk of sample attrition throughout the follow-up period.

The main consequence is the introduction of selection bias,

which contributes to uncertainty and unpredictability in the

outcomes.

Fig. 2 Interaction model and an example with public health inter- ventions, risk factors and diseases

462 F. Attena

123

From the above the last well-known issue is derived.

Difficulty in applying the RCT design

The wide discussion about this issue (Campbell et al. 2000;

Rychetnik et al. 2002; Dobrow et al. 2004; Hawe et al. 2004;

Victora et al. 2004; Petticrew et al. 2012) addresses the fol-

lowing questions: because the RCT design was originally

developed for interventions that are independent of context, it

is necessary for the public health intervention to incorporate

contextual information into the design of the RCT (Pickett and

Pearl 2001). Similarly, it has been suggested that in addition to

the outcome evaluation, a process evaluation should be

included. A process evaluation can help distinguish between

interventions that are inherently faulty (because of concept or

theory) and those that are badly delivered (implementation

failure) (Oakley et al. 2006); another way is to achieve a high

level of standardization of the whole intervention (Craig et al.

2008). When it is impossible to carry out RCTs, i.e., when the

unit of intervention is a community or when it is very difficult

to implement randomization or double-blind conditions, it is

important to look at and evaluate other forms of evidence.

Returning to the characteristics of complex systems, we

have seen that public health interventions, particularly

health-promotion interventions, are composed of a large

number of interacting components and by an intricate web of

causation. These components can produce self-organization,

retroaction, and emergent properties owing to human inter-

action between delivery and receipt of the intervention;

finally, they are poorly predictable for the above character-

istics, mostly because of context dependence.

Four typologies of evidence

As we have seen, starting from the causal chain when a

public health intervention is performed, it is possible to

carry out two levels of evaluation: the effect of the inter-

vention on risk factor reduction (first relation), and the

effect of risk factor reduction on disease reduction (second

relation). Each of these relations may be evidence-based or

not according to previous knowledge or the condition of

self-evidence. When a relation is evidence-based or self-

evident, it is useless in making an evaluation. On this basis,

the opportunity to evaluate neither, one, or both relations

arises (Table 2).

Evidence of both relations

Much environmental epidemiology is included in this cat-

egory. For example, water disinfection ? microorganism elimination ? infectious disease reduction; application of filters ? reduced dioxin intake in the atmosphere ? can- cer reduction. In both these examples, it is clearly unnec-

essary to carry out any evaluation because the causal chains

are evidence-based through the contribution of current

knowledge; in the second example, the first relation (fil-

ters ? reduced intake) can even be defined as self-evident. This typology is the only one in which it is possible to

speak of evidence-based public health.

Evidence of the second but not of the first relation

Evidence only of the second relation occurs typically in

health-promotion interventions. For example, health edu-

cation interventions to reduce smoking habits ? smoking habit reduction ? decrease in lung cancer incidence. The weakness of the first relation depends on the high context

dependence of health education interventions. In fact, we

may never be sure that a successful intervention in a given

situation is successful whenever and wherever. Instead, the

second relation is ensured by risk factor epidemiology. In

such cases, it is sufficient to check the reduction of risk

factors as a good result of the preventive intervention.

Table 2 The four typologies of evidence in public health interventions

The four typologies of evidence

I II III IV

First relation Evidence-based Grading of

evidence

Evidence-based Grading of

evidence

Second relation Evidence-based Evidence-based Grading of evidence Grading of

evidence

Evaluation of risk factor reduction No Yes No Yes

Evaluation of disease reduction No No Yes Yes

Evaluation by local public health departments No Yes No No

Evaluation by experimental research centers No Yes Yes Yes

Prevalent typology of public health

interventions

Environmental,

clinical

Health promotion Environmental,

clinical

Behavioral

Complexity and indeterminism of evidence-based public health 463

123

Evidence of the first but not of the second relation

This category occurs when the risk factor involved is still

putative. For example: relocation of power lines away from

towns and populations ? reduction in extremely low-fre- quency electromagnetic field exposure ? childhood leu- kemia reduction. The first relation is self-evident, whereas

the second relation is still uncertain. In these cases, it should

be necessary to conduct an evaluation of the real reduction

in disease incidence after the preventive intervention.

No evidence of either relations

The lack of evidence of all causal chains would require an

evaluation of both relations. For example, health education

to the owners of food stores and restaurants ? improve- ment in sanitary conditions ? foodborne disease reduc- tion. It is known that health education does not always

change behavior and that improving sanitary conditions in

food stores and restaurants is insufficient to produce a

reduction in foodborne disease in the general population;

however, this second relation is a little more consistent than

the first. This typology also occurs when there is a lack of

previous studies and knowledge about the two relations.

This model can be integrated with other models that

evaluate and grade evidence in public health (Briss et al.

2000; Tang et al. 2008). Such grading can be applied

separately to both causal relations.

The four typologies are useful in distinguishing pre-

ventive interventions conducted in local public health

departments and in experimental research centers. With

minor expertise and resources, the former types of insti-

tution can apply the first and second typologies, whereas

the latter types are also able to apply the third and fourth

typologies, which involve more complex evaluations of the

second relation.

References

Attena, F. 1999. Causal models in conventional and non-conventional

medicines. Medical Hypotheses 53: 177–183.

Briss, P.A., S. Zaza, M. Pappaioanou, J. Fielding, L. Wright-De

Agüero, B.I. Truman, et al. 2000. Developing an evidence-based

guide to community preventive services–methods. The task force

on community preventive services. American Journal of

Preventive Medicine 18(Suppl. 1): 35–43.

Brownson, R.C., J.E. Fielding, and C.M. Maylahn. 2009. Evidence-

based public health: A fundamental concept for public health

practice. Annual Review of Public Health 30: 175–201.

Bunge, M. 1979. Causality in modern science, Third revised edition

ed, 17–19. New York: Dover.

Buring, J.E. 2002. Special issues related to randomized trials of

primary prevention. Epidemiologic Reviews 24: 67–71.

Campbell, M., R. Fitzpatrick, A. Haines, A.L. Kinmonth, P.

Sandercock, D. Spiegelhalter, et al. 2000. Framework for design

and evaluation of complex interventions to improve health. BMJ

321: 694–696.

Craig, P., P. Dieppe, S. Macintyre, S. Michie, I. Nazareth, M.

Petticrew, et al. 2008. Medical Research Council Guidance.

Developing and evaluating complex interventions: The new

Medical Research Council guidance. BMJ 337: a1655.

Dobrow, M.J., V. Goel, and R.E. Upshur. 2004. Evidence-based

health policy: Context and utilisation. Social Science and

Medicine 58: 207–217.

Galea, S., M. Riddle, and G.A. Kaplan. 2010. Causal thinking and

complex system approaches in epidemiology. International

Journal of Epidemiology 39: 97–106.

Haddix, A.C., S.M. Teutsch, P.A. Shaffer, and D.O. Dunet. 1996.

Prevention effectiveness in health and medicine, 8–9. New York:

Oxford University Press.

Hawe, P., A. Shiell, and T. Riley. 2004. Complex interventions: How

‘‘out of control’’ can a randomized controlled trial be? BMJ 328:

1561–1563.

Jackson, N., E. Waters, and for the Guidelines for Systematic

Reviews in Health Promotion and Public Health Taskforce.

2005. Criteria for the systematic review of health promotion and

public health interventions. Health Promotion International 20:

367–374.

Jenicek, M. 1997. Epidemiology, evidenced-based medicine, and

evidence-based public health. Journal of Epidemiology 7:

187–197.

Joffe, M., M. Gambhir, M. Chadeau-Hyam, and P. Vineis. 2012.

Causal diagrams in systems epidemiology. Emerging Themes in

Epidemiology. doi:10.1186/1742-7622-9-1.

Karhausen, L.R. 2000. Causation: The elusive grail of epidemiology.

Medicine, Health Care and Philosophy 3: 59–67.

Kemm, J. 2006. The limitations of evidence-based public health.

Journal of Evaluation in Clinical Practice 12: 319–324.

Kohatsu, N.D., J.G. Robinson, and J.C. Torner. 2004. Evidence-based

public health: An evolving concept. American Journal of

Preventive Medicine 27: 417–421.

MacMahon, B., and T.F. Pugh. 1996. Epidemiology: principles and

methods, II Edition ed, 26–29. Boston: Little, Brown and

Company.

Oakley, A., V. Strange, I.C. Bonel, E. Allen, and J. Stephenson. 2006.

Process evaluation in randomised controlled trials of complex

interventions. BMJ 332: 413–416.

Olsen, J. 2003. What characterises a useful concept of causation in

epidemiology?. Journal of Epidemiology and Community Health

57: 86–88.

Parascandola, M., and D.L. Weed. 2001. Causation in epidemiology.

Journal of Epidemiology and Community Health 55: 905–912.

Parascandola, M. 2011. Causes, risks, and probabilities: Probabilistic

concepts of causation in chronic disease epidemiology. Pre-

ventive Medicine 53: 232–234.

Pearce, N., and F. Merletti. 2006. Complexity, simplicity, and

epidemiology. International Journal of Epidemiology 35:

515–519.

Petticrew, M., Z. Chalabi, and D.R. Jones. 2012. To RCT or not to

RCT: Deciding when ‘more evidence is needed’ for public

health policy and practice. Journal of Epidemiology and

Community Health 66: 391–396.

Pickett, K.E., and M. Pearl. 2001. Multilevel analyses of neighbour-

hood socioeconomic context and health outcomes: A critical

review. Journal of Epidemiology and Community Health 55:

111–122.

Plsek, P.E., and T. Greenhalgh. 2001. Complexity science: The

challenge of complexity in health care. BMJ 323: 625–628.

Rickles, D., P. Hawe, and A. Shiell. 2007. A simple guide to chaos

and complexity. Journal of Epidemiology and Community

Health 61: 933–937.

464 F. Attena

123

Rychetnik, L., M. Frommer, P. Hawe, and A. Shiell. 2002. Criteria for

evaluating evidence on public health interventions. Journal of

Epidemiology and Community Health 56: 119–127.

Rychetnik, L., P. Hawe, E. Waters, A. Barratt, and M. Frommer.

2004. A glossary for evidence based public health. Journal of

Epidemiology and Community Health 58: 538–545.

Rychetnik, L., A. Bauman, R. Laws, L. King, I.C. Risse, D. Nutbeam,

et al. 2012. Translating research for evidence-based public

health: key concepts and future directions. Journal of Epidemi-

ology and Community Health 66: 1187–1192.

Susser, M. 1973. Causal thinking in the health sciences, 68–69. New

York: Oxford University Press.

Tang, K.C., B.C. Choi, and R. Beaglehole. 2008. Grading of evidence

of the effectiveness of health promotion interventions. Journal of

Epidemiology and Community Health 62: 832–834.

Tremblay, M.C., and L. Richard. 2011. Complexity: A potential

paradigm for a health promotion discipline. Health Promotion

International. doi:10.1093/heapro/dar054.

Victora, C.G., J.P. Habicht, and J. Bryce. 2004. Evidence-based

public health: Moving beyond randomized trials. American

Journal of Public Health 94: 400–405.

Complexity and indeterminism of evidence-based public health 465

123

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.