Methodology

Method and Work Plan

Utilizing a packaged simulation program, GPSS, on the University of Michigan's mainframe computer (the program is also available in a PC version), we have built a crude prototype ofthe intended simulation model, utilizing smoking cessation as the intervention of interest. At present, we have acquired some of the data we will ultimately need to run the model and are employing educated "guesstimates" for parameter values for which we do not yet have adequate data. The most basic data needs include the prevalence of smoking among workers of particular ages, the health effects and consequent costs associated with their smoking, the costs and effectiveness of intervention (cessation) programs, and the health implications of smoking behavior change. Such data are entered into the model, along with information reflecting business costs ranging from salaries to health care costs to costs of turnover, retirement, and so on. At present, we are employing broad, nationwide averages for the business activity and cost parameters. Eventually, we will want to use data specific to one firm or one class of firms (e.g., blue collar manufacturing).

We have run the prototype model with a hypothetical workforce of 2000 employees, examining effects over a 20-year period. The demographic characteristics of the employee population have been taken from national surveys, so that ourhypothetical firm represents the "average" of all firms nationwide. Once the model is running to our satisfaction, we will be able to substitute data better characterizing the composition of actual firms. In addition, we will want to examine larger workforces and, especially, to look at effects further into the future than the 20 years currently considered.

Attached to this project description is an appendix (Appendix A) including, in order: a simple flow diagram indicating the basic structure of the analysis; an outline of the specific steps taken in running the prototype simulation; a diagram of the simulation model (essentially a simplified presentation of the outline); a more complicated flow chart indicating the relationships we hope to be able to incorporate in the final model; five pages describing assumptions employed in the initial prototype modeling (some of the assumptions are based on actual data, while others are assumed for purposes of testingthe functioning of the model); two pages of illustrative tabular output; and several pages providing a few examples of model output in graphical representations.

The output presented in Appendix A should be interpreted as merely illustrative of what the model can produce, as considerable work remains to get the prototype to the point that we believe it will accurately represent what occurs in a firm that adopts a smoking cessation program. Nevertheless, the attachments demonstrate the nature of the basic output that the model will produce, ultimately to include smoking prevalence, health outcomes (reductions in smoking-related deaths and disability days), program costs, economic benefits (e.g., reductions in health care costs, disability costs, life insurance payments, absenteeism, and turnover costs), economic costs (e.g., increases in retiree pension obligations and supplemental health insurance), and demographic implications (age-sex mix of employees and retirees). Net economic implications and cost-effectiveness estimates will be calculated from these outputs (directly, as derived model output). Output will be available for individual years (e.g., outcomes during the fifth year) and over periods of time (e.g., cumulative outcomes through the fifth year). As noted above, output will be derived for both the firm and the community as a whole. (During development of theprototype, we examined only the firm-specific interests.)

Refining the prototype is the next stage of the research, work that, we estimate, should require from 12 to 15 months from inception of funding. The principal elements of this work will include:

Data acquisition, including the following categories of data:
- smoking prevalence by age-sex-occupational category
- health effects of smoking intervention
- participation rates, costs, and effectiveness
- health effects of cessation by length of time quit
- costs of smoking, including:
  - health care costs
  - absenteeism
  - life insurance costs
  - etc.
- structure and costs of business benefit plans, including:
  - employee and retiree health care
  - disability
  - pensions
  - etc.
- Workforce demographics
  - age
  - sex
  - occupational category
Refinement of the prototype model of the firm
- incorporate basic variables not now included
- permit easy modification of assumptions for purposes of sensitivity analysis.
Extension of the model to evaluate social externalities
- those health, economic, and demographic effects that "diffuse" outside of the firm but remain in the community (e.g., the health benefits accruing to former smokers who have left the firm).

The refined model will be fully functional and able to address directly a large number of the most fundamental questions associated with the consequences of a worksite smoking cessation program. The second phase of the research, expected to take approximately six months, will entail an attempt to expand the basic model to incorporate as many as possible of the more subtle relationships indicated on the second flow chart in Appendix A (page A-4), such as the direct effects of smoking cessation on on-the-job productivity, or the indirect impact on productivity and employee retention of the "good will" associated with the existence of an employee health promotion program.

Practically speaking, we recognize the difficulties we will encounter in grappling with such important, but elusive, issues as productivity. One of the principal virtues of this modeling approach, however, is the ability to employ sensitivity analysis liberally. This means that in situations in which parameter values are uncertain, for example, we can adopt a modal case and then, instantly, ask the model what will happen if we alter the value of the uncertain variable. This flexibility permits development of a range of findings in the case of variables for which we do not possess confident estimates. It helps to ascertain which uncertain variables warrant further investigation and which are relatively unimportant. In the latter case, it may tell us that important outcomes are insensitive to the value of the uncertain variable.28

The third and final phase of the research, occurring duringthe final three to six months of the grant period, will encompass three activities:

Testing and application of the model using illustrativeor actual data from specific firms (varied by size,type of employee population, etc.).
Preparation of research papers reporting results of the work. (We anticipate that this activity will be occurring throughout the second year. We refer here to completion of final papers.)
Exploration of the potential to extend the model to other worksite health-related interventions. Three illustrative candidate cases, based on their importance and the availability of epidemiologic and programmatic data, are:
1. alcohol abuse counseling and treatment (which might itself serve as a good model for a similar analysis of counseling and treatment of illicit drug abuse)
2. exercise programs
3. screening and treatment of hypertension or hypercholesterolemia.

Assuming that the smoking cessation simulation model works well -- that is, if it produces interesting, realistic, and useful policy-relevant findings -- the final step enumerated above will suggest whether it is worth pursuing similar work for additional interventions.